There are a number of known problems with IOCCC entries: many of which have to do with differences between today’s compiler environments and those of today. In some cases the original code has a bug, some of which were recently discovered and many of which were fixed. In some cases attempt to port older code to more modern environments introduced bugs.
The following is a known list of BUGS and (mis)FEATURES in IOCCC entries.
Can you fix/improve entries not under the INABIAF (it’s not a bug it’s a feature)? You are VERY WELCOME to try.
NOTE: currently this file is mostly for us. We are not yet ready for help except for dead links and missing files. We will remove this notice when this changes. We will note this below for each status we’re not yet ready for help with.
As far as how to help: please submit your fixes in a GitHub pull request (with ONE PULL REQUEST PER FIX, please)!
We will be happy to credit anyone who submits successful GitHub pull requests in the thanks file. If you’re the author of an entry that win the IOCCC, we will add a link to your winning entries in the file (if you’re not an author, we can add a link to your GitHub page or personal website if you have one, should you wish).
If you do fix an entry please feel free to delete the entry from this file! Otherwise if you wish to not worry about it we can do that to make sure the format is consistent and clean.
THANK YOU!
If you’re the author of an entry that has been fixed and you find it against your liking PLEASE let us know and we’ll be happy to undo any fixes even if it takes away some instructional value or even usability. _In this case we’re VERY SORRY_ about it: it’s a fine line, we know, and we tried to use careful judgement but invariably some might have been the wrong decision. Thank you for understanding!
I (Cody Boone Ferguson) accept the blame in some cases; when working on getting entries to compile sometimes I got carried away with it and there might be that I fixed some things that should not have been fixed. I am correcting these mistakes as I come across them but if I miss any please let the judges or me know.
Make ABSOLUTE CERTAIN that you test the entry
BEFORE AND AFTER your changes! This
includes output and the same input functionality! Sometimes it might
seem to be fine but is actually not! We will note some where this is
known to happen (if it’s not yet fixed). This has actually already
happened to us. Sometimes it is okay: some entries are known to not work
but they can still have improvements (rarer situation but it’s
happened). Some might have slightly different output but which is not a
problem for instance a newline after output in Cody Boone Ferguson’s
changing it to 1994/ldb use
fgets() from gets().
Make ABSOLUTE CERTAIN that you read the README.md file BEFORE your changes as it’s important to see that the code is doing what it is supposed to. In the case that it’s not obvious (some README.md files do not even have commands to try or even explain how to use it!) then you should either skip the entry OR ask someone who will know e.g. us.
Make ABSOLUTE CERTAIN that you read the next section, the list of statuses and the related information, BEFORE you submit a pull request!
BE VERY AWARE that sometimes fixing an entry for one platform will break it under another so if you have more than one platform (e.g. macOS and linux) it would be wise to test it under all that you have access to. Otherwise we can do that. Of course if the entry does not work under any platform and you fix it for one that’s more than fine (and it has been done numerous times).
Again, THANK YOU!
NOTE: when going through the entries you can skip to the next entry
(some are detailed explanations) by searching for
^## STATUS: in your editor (or searching for it - probably
without the ^ in your browser). For a specific year try
^# YYYY where YYYY is the year as a
number.
Entries below have one or more of the following STATUS values. Please see the text below for more information. If an entry has more than one status it means that either they all apply or they compliment each other. For instance 2004/gavin crashes but it also doesn’t even compile with some platforms/architectures.
Cody Boone Ferguson has been working on resolving these issues and has fixed many but you’re welcome to have a go at it too (though he’ll be very sad if he doesn’t have a chance to at first fix it :-( so you might want to hold off for now :-) .. or not :-) ). He hopes to have most of the remaining entries resolved in the near future but nevertheless if you’re okay making people very sad you may have a go at the entries :-) Yusuke Endoh has also fixed a number of entries though at this point he does not know it. :-) Cody will remove this part later on.
In general warnings should NOT be addressed. The only time they should be CONSIDERED is when the entry does not work. However note that sometimes trying to fix the warnings will actually introduce bugs! Other times ‘fixing’ them will break the entry (see below).
Below in some entries we do list some warnings that definitely should be ignored (including some introduced by fixes) but we do not list them all: trying to keep track of them all would be impractical especially as different compilers give different warnings.
Another type of warning that would be hard to keep track of is
different data sizes on different platforms. These tend to be required
at the risk that sometimes the entry will not work for certain
platforms, some of which might or might not be fixable; a good example
where it was required to change and is okay is when Cody Boone Ferguson fixed
the segfault in macOS of 1989/paul:
changing the int * to a long * was required
and it works just as well with linux.
But even if they are fixable (which will likely be hard to do) it’s almost certain that such code would be just as non-portable (importable ? :-) ).
In rare cases they are valid. For instance Cody noticed in linux the following warning in 1985/applin with gcc:
applin.c:1:78: warning: incompatible implicit declaration of built-in function 'execlp' [-Wbuiltin-declaration-mismatch]
1 | main(v,c)char**c;{for(v[c++]=strdup("Hello, world!\n");(!!c)[*c]&&(v--||--c&&execlp(*c,*c,c[!!c]+!!c,!c));**c=!c)write(!!*c,*c,!!**c);}
| ^~~~~~and this reminded him of his fix to 1984/anonymous. In the case of
1984/anonymous it was a bit more involved but with
1985/applin one need only add to the Makefile
-include unistd.h. So there are some cases where fixing
warnings can fix a problem but in general they should be ignored even if
they’re annoying.
Hopefully with the example entries listed above you get the idea.
If you’re fixing an entry please make as FEW CHANGES AS POSSIBLE! This is to make it as close to the original but allowing it to work. This might be less of a problem when providing alternate versions but it might still be nice to have it as close as possible to the original. See also below two points.
For one-liners PLEASE KEEP THE FILE ONE LINE IF AT ALL
POSSIBLE! If it needs an include you can update the Makefile
CINCLUDE variable. For instance if it needs
stdio.h you could do -include stdio.h. Please
leave a space after the = in the Makefile. You may also
have extra long lines if this seems useful to make it a one-liner even
if it kind of makes it longer than what the judges consider a one-liner.
Thank you!
If you make changes PLEASE TRY AND KEEP THE SOURCE CODE LAYOUT AS CLOSE TO THE ORIGINAL AS POSSIBLE. This might not always be possible and if you have an editor that does formatting it can cause problems. Sometimes formatters can even break code! Cody Boone Ferguson has experienced this many times with vim so he tends to disable all format options when formatting code. In vim you should be able to do that with:
:set formatoptions=NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Entries with this status have one or more bugs that need to be fixed. Are you able to fix it? We welcome your help!
An important point to note is that there is a very fine line when fixing bugs where it might sometimes border on tampering with the program. And after all, these are not meant to be maintainable or even good programming style! Use careful judgement when fixing bugs please!
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Entries with this status might or might not have a bug possibly depending on the system. In these entries it’s unknown if there is a bug and sometimes it’s because we do not remember and sometimes we don’t have the appropriate system or environment to test and fix any possible problems.
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Although these entries appear to work for one or more reasons we’re unsure if they are completely functional. Can you confirm this? Please let us know so we can fix it!
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Entries with this status almost certainly have a bug or some other problem. The issue or issues might depend on the system much like the above STATUS: possible bug (possibly depending on system).
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Entries with this status do not work under some OSes and/or architectures (and/or something else?). Please help us to fix it!
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Some entries do not work with some compilers. A good example is 1992/lush which uses error messages from the compiler to generate its output.
If you can provide code that works for multiple compilers without too much tampering this is okay though it’s probably not possible with the above mentioned entry; otherwise it would be better to provide alternate code. In some cases it might be better to provide alternate code anyway. Use a judgement call here as best you can manage.
NOTE: it appears that most if not all of these have been fixed except perhaps for 1989/westley but this has probably been fixed as much as possible given the nature of how it generates code: the entry itself compiles but two versions of code it generates does not work with clang.
However some still might have problems of some kind or another and
some might not yet be located as an additional defect of clang was
noticed where main() is not allowed four args or one
arg.
This is what the status means, however.
Entries with this status have a problem in that the args to main()
are not of a specific type or there are too many due to this being
allowed in earlier C. Some compilers like clang have a defect where they
do not allow this and some versions have an additional defect where they
only allow 0, 2 or 3 args, the latter of which affected
1989/westley, so these entries do not work with clang.
Cody Boone Ferguson
looked at the source code of clang and reported that there is no way to
override the requirement of arg types so these entries will fail to
compile with clang. That’s why in some entries he fixed he did it by
adding a function called pain(). :-) He did not check for
the number of args as that was discovered later but he suspects it’s the
same issue.
NOTE for macOS users: please be aware that
gcc under macOS is actually clang despite the
fact it might appear to be gcc: no symlink and both gcc and clang exist
but the gcc is clang which you’ll see if you run
gcc --version.
A tip and some fix methods from Cody: in the older days args to
main() not given a type were implicit ints but when they’re required to
be char ** this can cause a problem. In some cases Cody was
able to use a char * inside main() (see 1989/tromp/tromp.c and 1986/holloway/holloway.c for two examples
though done slightly differently). In other cases he was able to
dereference the pointers to be used like an int and other times a cast
was necessary. He used other various techniques to get them to compile.
In some cases this introduced a problem but typically if not always that
problem exists with compilers that are less strict.
Because some versions of clang complain about the number of args of
main() and in particular say (when it has four, for
instance) that it’s only allowed to have 0, 2 or 3 args, numerous
entries had a second arg added to main() (if they only had
one). This is in case clang adds further restrictions which is not
impossible especially as it already claims having only one arg is not
allowed. At this time (04 Feb 2024) there is one entry known (it is
possible that not all were checked) that has this problem but it’s not
as simple as the others to fix without breaking it.
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
While such entries can compile, the resulting executable sometimes or always crashes.
NOTE: this does NOT apply to entries under the INABIAF status (though they’re not mutually exclusive in some cases).
REMINDER: if you’re debugging a crash it will be very helpful to have
-O0 -g or if you can -ggdb3 when compiling as
that will help with debugging symbols.
NOTE: we’re still locating entries and working on fixes with this status so we’re not yet ready for help. We will remove this when we are.
Entries with this status use gets() which is unsafe
because it has no limit on the length of the string. Cody Boone Ferguson has
fixed several of these and he’s working on others in between fixing
other things.
There’s a very good reason for this even though it is bordering on tampering entries: in modern systems one will get possibly alarming and certainly annoying warnings during at least one of compiling, linking and execution. In macOS it’s most annoying as it happens during execution (also compilation) with the warning being interspersed with the output of the program, often causing confusing output with the entry. A good example that was fixed is 1990/tbr.
Cody provided some important notes on this status with respect to fixing / changing entries:
In some cases changing the entry to use fgets() will
break the entry and in some cases cause it to crash. Where this is known
we will document it. Some of these Cody is also off and on working
on.
You MUST check the output before and after to make sure that it remains the same. Sometimes the output might not be immediately obviously wrong but is. We will document known examples as we come across them.
gets() does NOT STORE the '\n' but
fgets() DOES! This is one of the reasons some of the
entries break when changed to fgets(). Cody has fixed some
of these but there are others he hasn’t had time to address or even
discover yet.
Just because you don’t see the string gets in the
code does not mean it’s not used. It is after all the International
Obfuscated C Code Contest! :-) Sometimes it will be in the
Makefile and other times it will be obfuscated in other ways. Compilers
and linkers tend to warn about its use (and as noted in macOS it also
happens at execution) and this is a good way to find entries that use it
even if it’s not visible in the code.
NOTE: this status is NOT necessarily mutually exclusive with the INABIAF (it’s not a bug it’s a feature) status. The reason for this is due to warnings during compiling, linking and/or runtime, sometimes causing confusing output (as noted above).
Sometimes getline(3) will work but note that this
function also stores the newline just like fgets(3).
In these entries one file or multiple files are missing from the repo. In some cases these files can be found on the IOCCC website but in other cases they are entirely absent. In this case you’ll probably have to contact the author (unless you are the author! :-) ).
In these entries one or more make rules noted by the author and/or judges is/are missing. You might consider looking at the archive as a good place to find them.
This is as they sound: a link is either missing or it’s no longer valid. In many cases the Internet Wayback Machine will be very useful but there happens to be numerous links that this is not helpful. In other cases the URL has changed. Some of these have been discovered by the Internet Wayback Machine with the orange status.
The archive website will tell you if the link was never captured.
Entries with this status should NOT be touched (unless they have another status that suggests that that issue can be changed): they are system specific by design or are other things that are not actually bugs or bugs even if they appear to be. Or they might be documented bugs or things that simply are part of the entry. Nonetheless they are noteworthy.
NOTE: the definition of this status changed over time. At one point it was when something was noted by the author, the judges, a consequence of earlier requirements for winning entries or the purpose was to do something that might appear to be buggy. An example of system specific entries:
1984/mullender (see below and the FAQ for a version that works in modern systems) is very system specific and was before system specific winning entries were discouraged. This is an all time personal favourite of Landon Curt Noll. Run the alternate code to understand why this might be (along with how strange the source code is :-) ).
An example where a crash is not a bug: 2019/endoh is supposed to crash. There are others that are also supposed to crash or that are known to segfault but are considered features.
An important note is that if the README.md of the entry has a bug status that says it can be fixed it can be. Otherwise it should not be.
Nonetheless we challenge you to fix these entries for educational/instructional value and/or enjoyment but we kindly request that you DO NOT submit a pull request unless it’s a bug or (mis)feature we would like you to fix! If you can’t figure it out you’re invited to look at the git diffs, where there are some (some were fixed earlier on but rolled back as both Cody and Landon individually felt that the fix was tampering with the entry).
NOTE: in the case of gets() we’ve fixed some to avoid
the warning of the compiler, linker or even during runtime, depending on
the system. In some cases like 1990/tbr the
fix actually prevents confusing output (though that was not the only fix
made in that entry).
Of course if you’re the author you’re welcome to fix your own entry, prefer your own fix or suggest that they’re fixed!
The purpose of this program is to print out a string of rubbish. In particular you should see something like:
$ ./decot
'",x); /*
\without a newline after the \. This is not a bug.
Although there is an alt version and supplementary program added by Cody that will work in modern systems, if you do not have a VAX-11 or PDP-11 to run the original entry on it will not work. See the README.md for details on the alternate versions.
Cody added and fixed the gentab.c which is from the author’s
(or one of them, Mullender) remarks found by Cody. Cody fixed this to
compile and work (as best as he can determine: he has no VAX-11 or
PDP-11 or emulator to test it) but running the code on the binary itself
produces a short[] that can compile in modern systems
though it’ll not work.
There are no known bugs and (Mis)features for entries in 1985.
This entry is known to segfault after printing its output. It was documented by the judges and shouldn’t be fixed.
There are no known bugs and (Mis)features for entries in 1987.
In linux it might happen that despite no error message or message about doing so, the program drops a core file into the directory even though the entry works and does not crash.
If you use either fubar or ouroboros.c
(it’s executable, see README.md for details) with a number < 0 or
larger than, say 20, it’s very likely that the program will turn into an
infinite loop trying to compile code with syntax errors.
If you want to try and fix this (mis)feature, you are welcome to try.
This program will very likely crash or break into tiny bits :-) if you feed it numbers with non-binary digits.
There are also other cases where this can happen for instance using
unsupported operators like /. To see what operators are
supported check the source code.
With version 2 it sometimes segfaults even with the same input where other times it does not. We don’t believe this is because of the fix that lets some versions be compiled with clang. An example invocation is:
./ver2 < westley.cIt should be noted that in additional to rot13 names there is code that is the reverse of other code (also wrt names). See the source file and the README.md (in the author’s remarks) for more details.
Fixing the (Mis)feature is likely to be a very difficult challenge especially without breaking something else which is far more likely (see below in tips from Cody). You are welcome to try and fix it if you can!
The reason this is crashing is that the array irk is
being accessed way out of bounds by the int gnat. For
instance:
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x0000000000401335 in main (ABBA@entry=<optimized out>, tang@entry=<optimized out>, gnat@entry=<optimized out>, Near@entry=<optimized out>)
at ver2.c:23
23 &&gnat!({)Near,noon,/*krelc*/)<0&&(Near= - irk[-gnat--]-2)))&&main(ABBA,
(gdb) p gnat
$1 = -518733305The real trouble is that the code is generated and in a complex way or rather ways.
I will explain this a little bit (though it might be more than a bit according to many :-) ) but that is all I can do for now. The program, when used without any args, will print out what it reads. If one arg is specified it will do a ROT13 of it. If two args are specified it will print out the text backwards (WITHOUT ROT13!). If three args are specified it will do both reversal AND ROT13.
So ver0 is the same as the main program;
ver1 is the ROT13 of the code (but see below);
ver2 will reverse the code and ver3 will both
reverse and ROT13 the code. But it’s trickier than that as one might
expect.
The comments are processed too. Furthermore it happens that the last line becomes the first and the first line becomes the last! Now if you look at the code you’ll see on the first line:
/**//*/};)/**/pain(/*//**/tang ,gnat/**//*/,ABBA~,0-0(avnz;)0-0,tang,raeNIn ver0 it’ll be the same but the others are more
interesting.
In ver1 you’ll see on the last line:
cnva((vag)NOON&2/*//*\\**/,gnat,tang ,NOON/**//*/(niam/**/);}/*//**/and on the first line:
/**//*/};)/**/cnva(/*//**/gnat ,tang/**//*/,NOON~,0-0(niam;)0-0,gnat,enrAObserve that the ROT13 of pain is cnva and
the ROT13 of main is znva. The ROT13 of
NOON is ABBA. vag is ROT13 of
int, gnat and tang are ROT13 of
each other as well which is the reverse of the original (this is not
because it goes backwards (it doesn’t in this version) but because they
are ROT13 pairs!). main spelt backwards is
niam and pain spelt backwards is
niap. Furthermore see the Makefile for other defines that
had to be specified and for ver2 and ver3 (that both work with gcc) one
had to be undefined (search for CDEFINE and
-U). These details will be important momentarily.
As far as NOON and ABBA being ROT13 pairs,
notice how NOON is not in the original code but
ABBA is; but in other versions it becomes NOON
and they also have abba which the original code does
not!
Now notice how the avnz in the first line got changed to
niam, its ROT13 value. Notice also how some of these are in
comments!
main() is in there somewhere and that had to be changed
to call pain() (see the thanks file (see the
1989/westley section for details on how as this was not as
straight forward as it is for other entries) as well. Now if you notice
on that line you have in a comment niam. If you were to
change that to niap (pain backwards) that function would
end up as pain() in generated code! That’s assuming that
it’s just the reversal version, of course but ver1 is the ROT13
version.
These facts are bad enough but then you throw in the reverse of it without ROT13 and then another version that has the reversal and ROT13 then you can see it’s nigh impossible to fix if not impossible.
The fix to get it to even compile with clang and then get the first
two (counting the original, ver0) to work was the hardest
fix I made due to how the comments in reverse and ROT13 of it all works
together, especially allowing all versions to compile with gcc. The only
reason that it doesn’t work completely with clang is the arg types of
main() (see if you can figure out how the change of main()
in westley.c is not carried over to versions 2 and 3!);
originally none of them compiled with clang.
But it was possible to get the ROT13 code to compile with clang but the other versions, reversed as well as ROT13 and reversed code, proved much more problematic. I got it to compile (or maybe mostly compile) all versions with clang but I then carried it over to gcc in another system and it had syntax errors.
That’s about all I can say for how it works as other things have to be done too. Enjoy! :-)
A point worth considering is that as the number passed into the program gets bigger the number of lines of output gets substantially larger and it takes much more time and resources to run as well. For instance:
$ for n in $(seq 1 18); do echo "$n"; echo "$n" | time "./baruch" | wc -l | sed -E -e 's/[[:space:]]//g'; done
1
3
./baruch 0.00s user 0.00s system 5% cpu 0.115 total
2
1
./baruch 0.00s user 0.00s system 64% cpu 0.004 total
3
1
./baruch 0.00s user 0.00s system 60% cpu 0.003 total
4
11
./baruch 0.00s user 0.00s system 64% cpu 0.003 total
5
61
./baruch 0.00s user 0.00s system 66% cpu 0.003 total
6
29
./baruch 0.00s user 0.00s system 61% cpu 0.003 total
7
321
./baruch 0.00s user 0.00s system 69% cpu 0.002 total
8
829
./baruch 0.00s user 0.00s system 76% cpu 0.003 total
9
3521
./baruch 0.00s user 0.00s system 87% cpu 0.005 total
10
7965
./baruch 0.01s user 0.00s system 95% cpu 0.014 total
11
32161
./baruch 0.05s user 0.00s system 98% cpu 0.053 total
12
184601
./baruch 0.29s user 0.00s system 99% cpu 0.294 total
13
1031969
./baruch 1.78s user 0.00s system 99% cpu 1.794 total
14
5483941
./baruch 11.26s user 0.02s system 99% cpu 11.311 total
15
36466945
./baruch 79.20s user 0.15s system 99% cpu 1:19.72 total
16
251132705
./baruch 592.89s user 0.77s system 99% cpu 9:55.34 total
17
1724671873
./baruch 4535.71s user 5.25s system 99% cpu 1:15:49.76 total
18
12655721857
./baruch 37496.97s user 58.50s system 99% cpu 10:27:48.29 totalCody
Boone Ferguson fixed some issues in this program and Yusuke Endoh
provided the btoa script but it appears there is a bug in
this entry. The judges wrote that to test the entry one can do:
echo "Quartz glyph jocks vend, fix, BMW." | compress | ./btoa | ./jawwhich should apply the identity transformation to a minimal holoalphabetic sentence.
But doing this shows instead:
echo "Quartz glyph jocks vend, fix, BMW." | compress | ./btoa | ./jaw
a얖?)V...?????777??hC??h??-? )SSSSXXX????r?L=???*?-???ppp,,,?R
?j
111-)))? '..F@E
???b111?
..F..F.?n,,,,,L@E$Notice how there’s no newline at the end: that final $
is the prompt.
If one does, however:
echo "Quartz glyph jocks vend, fix, BMW." | compress | ./btoa | ./jaw | ./jawthey will get just
oopswhich seems to be an error message (one of the fixes was to make it
not use perror(3) - this fixed something else though it’s
no longer known what).
The script shark.sh has some issues
too in that due to path not having . (this and maybe some
other things were fixed) and tar not wanting to accept
reading from stdin (this in particular) even with the right
options used, seemingly, it has to write to disk the tarball which seems
to defeat the purpose. This would ideally be fixed.
The authors provided a list of features in the BUGS section in their remarks.
Cody Boone Ferguson
fixed many bugs that prevented this from working properly (including
segfaults) but one thing to note is that if you pass two zeroes to
theorem_bkp or fibonacci the program will
enter an infinite loop, printing 0 over and over again; another
condition where this occurred was fixed but this one should not be
fixed. Thank you.
If the maze file cannot be opened, either because the path specified does not exist or because the default (whatever the source file was at compilation time) file does not exist in the directory, this program will very likely crash.
This is a feature, not a bug.
There is a very simple way to always win. The program doesn’t catch you and as someone called Cody’s late grandmother said to him: ‘it’s not cheating unless you’re caught’. :-)
Please don’t try and fix it as it’s not a bug and was actually documented as a possibility. Can you find out how? There’s also a way to make it so that even when you’re cheating it ends up winning! Can you figure that out as well?
The author stated that if the file cannot be opened then it will print a system error but this is not the case unless it’s showing a fault. However this was instead fixed by Cody as part of another problem that was detected so that the program just exits 1 if the file cannot be opened.
Also, if some of the programs are not given an arg they will
very likely crash or do something funny. For instance
adsleep and adbasename. This is very simple to
fix but the author explicitly noted that this will segfault and it is
that that is the error message.
Another thing is that the code:
printf((W,Y));appears to be wrong because W is a FILE *
and Y is a char[] but if one changes it to use
fprintf() on the file (which, incidentally, is opened in
read only mode which is another reason) with the %s
specifier it will make adhead not work: it’ll print nothing
at all! So this code should not be changed either even if it appears to
be wrong. Notice too a curious thing: if you did change it to fprintf,
even if you have the right number of args, you’d have to remove the
outer () pair.
Leo Broukhis, before he was an IOCCC judge, sent the IOCCC judges an email:
From: leo _at_ zycad -dot- com (Leo Broukhis)
Date: Tue, 30 Jan 96 17:37:51 PST
To: judges _at_ toad -dot- com
Subject: IOCCC 1992 - a bug
Dear Judges,
albert.c (even in its fixed form) still has a bug. Although I don't
remember the number that exposed the bug (afair, resulting in coredump)
in albert.orig.c that has been fixed in albert.c,
I've found a number exposing another bug: 10000000001 (that's 9 0's).
Both albert and albert.orig loop without printing anything, although
the first factor is 101 and is usually found in an instant.A quick debugging session from Cody suggests that the problem with this value is that:
if ( ppp->qq!=48 ) return;is never reached because the value is never 48 (it is in other cases but not for the example input). It varies in value at this point. Observe that the next lines of code are:
while ( ppp->qq==48 )
{
printf("%ld\n",qqq-45);
*pp = ppp;
ppp = ppp->p;
}so it is expected that that value is 48, before printing the numbers,
but if it’s not it should not proceed at all, not even to go past that
loop. In other words that loop should always be entered for at least one
iteration. The check for the != 48 is also required.
Perhaps more useful to know is that at this point in the code the code:
if ((((( !(aa=setjmp(ppp->ppp))||aa==aaaaaa )))))appears to be entered because of the return value of
setjmp(), not the other condition.
The alt version does fix the problem but it is not obfuscated and not like the entry itself. Can you fix the actual entry? You are welcome to try and do so.
Cody changed it so that the buffer size is ARG_MAX+1 to
try and get past the problem of gets() being used in a more
complex way.
It would be ideal if it were to use fgets() though. This
will probably be looked at later but you’re welcome to try and fix this
too! A tip on how gets() is being used from Cody:
Previously it had a buffer size of 256 which could easily overflow.
In this entry gets(3) is used in a more complicated way:
first m is set to *++p in a for loop where
p is argv. Later m is set to point to
h which was of size 256. gets(3) is called as
m = gets(m)) but trying to change it to use
fgets(3) proved more a problem. Since the input must come
from the command line Cody changed the buffer size to
ARG_MAX+1 which should be enough (again theoretically)
especially since the command expects redirecting a dictionary file as
part of the command line. This also makes it possible for longer strings
to be read (in case the gets(3) was not used in a
loop).
When you start the program everything starts to move over to the right side and then ends. Yusuke Endoh pointed out that if you click the mouse it takes it back towards the centre.
If you want to try and fix this (mis)feature, you are welcome to try.
We used a patch from Yusuke
Endoh to get this to work but it only works with gcc. Cody removed
the warnings of gets() and provided a script to run the
entry properly.
Unfortunately due to the way the entry works and the fact that other compilers like clang have different warnings and errors this simply does not work with them.
Some tips from Cody:
This entry relies on specific compiler warnings. With gcc it will look something like:
[...]
lush.c: In function 'main':
lush.c:40: warning: "f" redefined
40 | main
|
lush.c:1: note: this is the location of the previous definition
1 | #define gets(_) fgets((_),999,stdin)
|
lush.c:42: warning: "f" redefined
42 | #define f 001:
|
lush.c:40: note: this is the location of the previous definition
40 | main
|
lush.c:44: warning: "f" redefined
44 | #define f 100:
|
lush.c:42: note: this is the location of the previous definition
42 | #define f 001:
|
lush.c:45: warning: "f" redefined
45 | _[i--
|
lush.c:44: note: this is the location of the previous definition
44 | #define f 100:
|
lush.c:46: warning: "f" redefined
46 | #define f 126:
|
lush.c:45: note: this is the location of the previous definition
45 | _[i--
|Now one can use sed to get the warning:
line correct but there is more to it than that. For instance this is
what it looks like with clang:
lush.c:40:9: warning: 'f' macro redefined [-Wmacro-redefined]
#define f 001:
^
lush.c:1:9: note: previous definition is here
#define f 000:
^
lush.c:42:9: warning: 'f' macro redefined [-Wmacro-redefined]
#define f 100:
^
lush.c:40:9: note: previous definition is here
#define f 001:
^As you might see the part under the warning: line is
different.
The entry is supposed to show warnings and then print:
Hello World.Can you help us?
When your own checkmate is imminent it prints "Har har"
but does not exit so it can ‘rub your nose in defeat’, as the author
puts it. You will have to exit it yourself through ctrl-c or killing it
in some other fashion.
Cody improved the usability of this program by making it so that as long as the terminal columns is >= 80 it will display properly, rather than having to wrap at 80 columns. However due to the nature of the program if the terminal is < 80 in column width it will not display right. To see the number of columns in your terminal try:
echo $COLUMNSThe author suggested that the alternate version, which Cody added (but fixed for modern systems), should print
main(l,a,n,d)...
[A M E R I C A]...…going down the left edge if your terminal auto-wraps at 80 characters.
This however does not seem to work. It does not appear to be because of the fix for modern systems as it doesn’t work for gcc either, even before any changes. This might be a misunderstanding or it might be that having 80 columns (resized) does not count (though the output before the fix of the 80 column requirement showed was correct when at 80-columns so it seems like this could be a bug if not a misunderstanding).
The author stated that:
The expression `(a*)*` compiles but loops forever.
There is no check for trailing backslash (`\`) in the pattern.
There is no check for unbalanced brackets. Omitting a closing bracket
will generate a "Pattern too long" error, which is not the real error.The author wrote in the documentation file ant.txt that there is a version of the code
that is not obfuscated. This file, agag.c does not exist in
the archive. Do you have it? Please provide it!
In multiple platforms, both macOS and also linux (in particular a RHEL 9.3 system), this entry just shows a blank screen.
Can you fix it? We welcome your help.
This entry relied on a bug in gcc that was fixed with gcc version 2.3.3. This cannot be fixed for modern systems as the bug is long gone.
An alternate version that will work for modern systems, however, does exist. See the README.md file for details.
Although the code checks if the file can be opened or not, badly formatted files will cause problems. No other checks are performed either.
There is no end of game checking method so you will have to quit the game through ctrl-c or such.
If the file cannot be opened it will very likely segfault. This should not be fixed except for an exercise to yourself, should you wish to try.
This program uses system(3) so if you provide invalid
characters in the question you might cause an error. For instance don’t
do this:
$ ./schnitzi schnitzi.info
This program answers questions about paragraphs
of text posed to it in English. It is written
in C. Mark Schnitzius is the author of this
program.
The Loch Ness monster has been captured. Elvis
is alive. Hostler was a woman. The NASA moon
landings were clearly faked.
? What is foo'?because doing so will result in something like:
sh: -c: line 0: unexpected EOF while looking for matching `''
sh: -c: line 1: syntax error: unexpected end of filewith the program terminating. Other characters will also cause this problem.
Of course if you do something like:
What is 'foo'?it will work fine.
No spaces are allowed in the expression.
The program does no error checking so erroneous expressions will produce spurious results.
The unary - is an operator so decimal -46
should be entered as -d46 and not d-46.
When you initiate shooting via the s command you
immediately lose an arrow before you choose where to shoot.
When you shoot it will move you to that room so if you end up shooting into a pit room you will end up dying even though you didn’t explicitly move there.
Cody Boone Ferguson
fixed this to compile with modern systems (see the thanks file for what had to change) but
the entry also used gets() which in some systems would
print out a warning along with the output of the program. Naturally it
could also overflow long lines. Cody changed it to fgets()
to prevent the display problem but this introduces another problem
namely that newlines can be printed if the line length < 231.
This seems like a worthy compromise to not have messed up output and although it would be ideal for it to never print a newline unless that’s the line itself it might be tampering too much with the entry as it’s not a real problem and as a one liner it’s already quite long.
NOTE: the generated code of all versions, when fed its own source, will differ even when it works. See the author’s remarks in the README.md for details. Increasing the buffer size and having it, when fed its own source code, generate code that will compile with the same buffer size is difficult: it just uses the original size. Cody explains this further down.
Getting it to use fgets() is even harder as when fed its
own code it will generate code that cannot even compile let alone use
fgets(3).
In both cases, changing the buffer size and changing it to use
fgets(3), will cause compilation errors without other
adjustments.
See the below magic for details. Along with the author’s remarks in
the README.md file it might prove possible to get it to use
fgets(3). Cody, writing this on 02 November 2023, only just
noticed the author’s remarks and will later on look at this if nobody
takes up the challenge. More important work like getting to a place that
the next contest can run must be done first.
Cody Boone Ferguson explains the magic of how this entry works, which will be necessary if this entry is to be fixed, below.
You might wish to run the command:
make diff_orig_progto see what had to change for the buffer size, when looking at the below. Furthermore you will want to look at:
diff schnitzi.alt.c schnitzi.alt2.cas the fgets() version can compile it just can’t
generate compilable code (let alone using fgets()) when fed
itself (to reiterate, just changing the call to fgets(3)
does not mean it can compile and there’s a further problem in that
fgets(3) retains the newline whereas gets(3)
does not). Nevertheless looking at these commands will be of help to
understand how it works.
For the alternate versions the other functionality is unaffected.
The problem is getting the generated code to use fgets()
(once it even compiles which was easy to do) and also have the updated
buffer size be the same (which was easy to do too at least in the
original version). The generated output, when changed to
fgets() failed to compile and it also used
gets(3) and these are the bugs here.
The buffer size, when using gets() is still the same but
as noted above the original code has had a buffer size increase.
The real problem is with formatting the code. Take a look at the
interesting comment as well as the int r=0,x,y=0 at the top
of the file. If you look at each column go down that column you can see
how it spells out the code! For instance the first column looks
like:
i
n
c
l
u
d
e
<
s
t
d
i
o
.
h
>If you join the lines you end up with:
#include <stdio.h>If you look at column 25 which is the end of the word ‘mh111’ and you
go down to the next row you’ll see a 0 and if you go one row down
another 0. This is the buffer size, 100, for u! The column
to the left is the same for the t variable.
Thus it seems that in order to get the generated output correct one needs to provide the correct input in comments or possibly by rearranging some of the code (this was actually required to make the generated code compile at all when changing the buffer size, see below).
Getting this entry to use fgets() is easy but the
problem is you’re supposed to be able to feed the source to the program
and the output of that will be compilable. It however will create
compiler errors. So it’s not just changing the code to get it to use
fgets()! I (Cody) already did this before I noticed the other part which
is why I rolled that part back. I did increase the buffer size but that
only works on the original source.
To say just how sensitive this entry is: even a space, lack of space or a character, wrong character, lack of character or character in the wrong place can cause a compilation error! Make sure that the output of:
./schnitzi < schnitzi.ccan be compiled and the output of that new program when fed itself can also be compiled!
This program will likely crash if the source code file (by the name of the file that’s compiled) cannot be opened in the directory it is run from.
Cody Boone Ferguson noted that the README.md file refers to an alternative version of the code that is not obfuscated but it is missing from the entry directory and the archive. Do you have this file?
We would be grateful if you could provide it to us. If you can provide this file you might consider removing this entry from this file as well but if not we’ll take care of it.
-1 value check for
getc():Cody Boone Ferguson
fixed the code to not use -1 for the return value of
getc(); this is important because EOF is
NOT guaranteed to be -1 but rather any
negative value. On systems where EOF != -1 the program
would enter an infinite loop until the program crashed, by chance reads
a -1 or was killed (it is for this same reason that one
should not use EOF for the getopt() functions
as they return -1 when all options are parsed (for details
on the definition of EOF see
7.21 Input/output<stdio.h> subsection 1 of the
standard)).
An interesting problem occurred where changing the -1 to
EOF caused both
warning: illegal character encoding in string literal and
error: source file is not valid UTF-8. But since
EOF is simply an int < 0 and making the loop condition
check that the return value is >= 0 does not cause a compilation
error and it functions correctly it will address the systems where
EOF != -1 just as if it checked for
!= EOF.
Since it works there is no need to fix this except for a challenge to yourself.
The judges said the following in their remarks:
The fractally minded may be able to detect that mode 0 does not calculate
Mandelbrot/Julian sets correctly. Can you find the bug? Better still, can you
fix it without breaking something else?You are welcome to try and fix it and open a pull request, providing that it doesn’t break something else.
This did not originally compile under macOS and after it did compile under macOS, it crashed. Cody Boone Ferguson fixed these problems.
It should be noted however that there is a condition where the
program will prompt you to press return again. This was thought to be a
bug but looking at the code it can clearly be seen that if
g - a is 0 then the message is supposed to be printed again
and one is supposed to press a key as at that point it calls
getchar() via the pointer m. So this is a
feature not a bug.
The judges suggested that the following commands should result in output:
./leo 1 | cat - /dev/tty | gs -
./leo 37 80 | cat - /dev/tty | gs -The first one can work if one does instead:
echo "" | ./leo 1 > foo.ps
gs foo.psbut it is supposed to work in the pipeline as is. The second one does not appear to work at all even if one redirects to a file (like the workaround for the first) as it appears to just block.
It is not known if this is platform specific but this was observed in macOS and it would be good if it was fixed.
This program expects a POSITIVE number. If you specify a negative number it will not show any output, stuck in a loop.
The authors stated that they included a version that allows people with just K&R compilers to use the program but this file is missing. Can you provide it?
We would appreciate anyone who has it or even just knows the name! Thank you.
The link was http://www.tc3.co.uk/~gandalf/G.HTML but this no longer exists as it was instead requiring a login / password.
Do you have an updated link? We welcome your help!
The changing to using fgets(3) in this entry is a bit
more complicated. It can almost be done except that some of the output
of the try.sh is wrong, especially the
last one.
This diff almost does it but not quite:
diff --git i/1996/huffman/huffman.c w/1996/huffman/huffman.c
index a5745bc7f4fa28b834c004f4cf19633e40ad9165..5e5e5334f33f9dbd95c70eddece3189d9bcff5e9 100644
--- i/1996/huffman/huffman.c
+++ w/1996/huffman/huffman.c
@@ -1,15 +1,16 @@
#define x char
#define z else
#define w gets
+#define gets(b) ((b)[0]='\0',(fgets((b), 100, stdin)))
#define r if
#define u int
#define s main
#define v putchar
#define y while
#define t " A?B?K?L?CIF?MSP?E?H?O?R?DJG?NTQ?????U?V?????X???????Z????W??Y??"
- s ( ) { x* c , b[ 5 * 72 ]; u a, e , d [ 9
+ s ( ) { x* c , b[ 5 * 72 ]; u a, e , d [ 9
*9 *9 ] ; y (w ( b) ){ r ( 0 [ b] -7 *
5 ) { c = b ; y ( (* c - 6
* 7 )* * c )c = c+ 1 ; r (( -0 ) [ c ] && w ( b +
8 * 5* 3 ) && w (b +8 *5 * 6) )
{ a = 0; y ( a [ b ]But since it does not for the time being it is advisable to just
redirect stderr to /dev/null
(2>/dev/null).
If X is not running this program will very likely crash or do something funny. This should NOT be fixed.
NOTE: the two boards will be on top of each other so you will have to drag one off the other so that you can properly play.
As well: the link which was http://www.uio.no/~jonth is no longer valid and there’s no archive on the Internet Wayback Machine. Do you know of a proper URL? We greatly appreciate your help here!
There was no IOCCC in 1997.
The domain http://pootpoot.com no longer exists as it once did. The judges have given a script that can be used to make a similar page (warning: not checked for security in modern days!). Do you have a server with enough bandwidth and would like to set it up? We’ll gladly thank you in the README.md file and link to the page as well! You’ll have IOCCC fame for reviving a pootifier! :-)
See above entry 1998/dlowe.
A point worth considering is that as the number passed into the program gets bigger the number of lines of output gets substantially larger. For instance:
$ ./schnitzi 9|wc -l
771999The larger the number the bigger the file can become too, even
becoming gigabytes in size. The range is checked for
>0 && <27 as not having this can be a problem
including a segfault.
If you use the generated program and do not give enough numbers in input something funny will happen, very possibly with different results per run. This is in the README.md file as something to try and ponder.
There was no IOCCC in 1999.
This MIGHT be better described as doesn’t work with
some versions of perl: the author states that in perl < 5.6.0 there
is a bug with a core dump in what they said is in
Perl_sv_upgrade. This does not exist in the code,
seemingly, but it might be that this is impossible to fix without
getting a perl upgrade. In any event the below details a problem.
Another bug that happens in linux also is below.
In macOS this program crashes. A tip from Cody is that the code that crashes is:
perl_eval_sv
(newSVpv("_:$_= <> ; defined or exit; @ARGV"
"=split; __:$_= shift;defined or goto _; chomp "
";(m*^\\x70oO" "\\x74$*i)?(pri" "nt \"$_ w\\x"
"61s h\\145r\\x" "65!\\n\"):((m*" "^\\s\\*(-?\\d+"
"(\\.\\d+)?)\\s" "\\*$*)?(push@SS ,$1):(&{chr(((o"
"rd)%39)+3**4)x2 } )); goto __; sub ff { @SS= ("
")} sub __{print \"stack empty\\" "n\"} sub ss{$#"
"SS<0 and goto & __; print $SS[$" "#SS].\"\\n\"} "
"sub SS{ $#SS<0 and goto &__ ; print pop @SS}"
"sub _ { print \"divide by zer" "o\\n\"}sub ii{"
"map{ print\"$_" "\\n\" } reverse @SS} sub AUTOLO"
"AD { print\"un" "implemented\\n" "\"} sub gg{ $#"
"SS<0 and goto &__;push@SS,$SS[ $#SS]} sub uu{ "
"$#SS<1 and goto &__;$SS[ $#SS]+= $SS[$#SS-1];$SS"
"[$#SS-1]=$SS[$" "#SS]-$SS[$#SS-" "1]; $SS[$#SS]-"
"=$SS[$#SS-1]} ",0),0 );and in particular it appears that it is the perl itself. In other words if one changes it to be:
perl_eval_sv(newSVpv("",0),0);it will not crash. But of course it won’t do anything either. So it’s
more specifically that the call to newSVpv() crashes the
code but INSIDE (key point!) the call to
perl_eval_sv() that does it. Given what the author stated
and that the version of perl is < than 5.6.0 that might be exactly
what is going on.
The example command above should print:
$ echo "13 14 15 16 17 + - * / p" | ./dlowe
-0.0515873015873016and that’s what it shows in linux.
Note that with different perl libraries in macOS (for instance from Homebrew and MacPorts it will likely print different warnings when compiling. It was observed that with Homebrew it does not report any warnings but with MacPorts it results in a total of 92 warnings! Nonetheless neither works okay and both crash.
The author gives an example command:
echo "7 P 6 d P P 8 p" | ./dlowe | tr 876 tpowhich should print out poot but it doesn’t, not in linux
and not in macOS.
In linux it doesn’t crash but it doesn’t print anything out either.
In macOS it crashes; it might appear to not crash in macOS but this
is because of the pipeline. If you remove the | tr 876 tpo
part of the command you will see that it does indeed crash!
The reason for this not working is really strange.
$ echo "7 P 6 d P P 8 p" | ./dlowe
7668
$ echo "7 P 6 d P P 8 p" | ./dlowe | grep 7
$Why? Is it writing to stdout? Let’s try some other things:
$ echo 7668 | tr 876 tpo
pootOkay so we know that it SHOULD work. But we also know something funny is going on with stdout and the entry. Another experiment:
$ echo "7 P 6 d P P 8 p" | ./dlowe 1>&2 | grep 7
7668Okay so now it sees it, grep. But watch!
$ echo "7 P 6 d P P 8 p" | ./dlowe 2>foo 1>&1
7668
$ cat foo.. so at this hour it does appear to be writing to stdout but yet somehow it doesn’t? But watch:
$ echo "7 P 6 d P P 8 p" | ./dlowe 1>foo 1>&1
$ cat foo
$Well this explains why the tr does not transliterate it
to poot but why is this happening? Why can’t it be
redirected to another file even?
If it could be figured out why it’s not writing to stdout and yet at the same time is writing to stdout one bug could be fixed. Maybe it’s the perl messing with things but we don’t know.
Other commands do work, however, and give the appropriate output, such as:
$ echo 1 2 + p | ./dlowe
3so something is wrong with some commands/operators.
Another feature that does work in linux is:
$ echo poot | ./dlowe
poot was here!Can you help fix these problems?
Cody provided the C code that creates an unformatted perl script that
will just print ‘unimplemented’ for operators but it seems to work in
some other ways (see below for how to go further than unimplemented)
which is the exact perl from the C program except that it has the
#!/usr/bin/env perl and use warnings; as well.
A note below might suggest that doing this creates incomplete perl.
Anyway:
#include <stdio.h>
int main(void)
{
printf("#!/usr/bin/env perl\nuse warnings;_:$_= <> ; defined or exit; @ARGV"
"=split; __:$_= shift;defined or goto _; chomp "
";(m*^\\x70oO" "\\x74$*i)?(pri" "nt \"$_ w\\x"
"61s h\\145r\\x" "65!\\n\"):((m*" "^\\s\\*(-?\\d+"
"(\\.\\d+)?)\\s" "\\*$*)?(push@SS ,$1):(&{chr(((o"
"rd)%%39)+3**4)x2 } )); goto __; sub ff { @SS= ("
")} sub __{print \"stack empty\\" "n\"} sub ss{$#"
"SS<0 and goto & __; print $SS[$" "#SS].\"\\n\"} "
"sub SS{ $#SS<0 and goto &__ ; print pop @SS}"
"sub _ { print \"divide by zer" "o\\n\"}sub ii{"
"map{ print\"$_" "\\n\" } reverse @SS} sub AUTOLO"
"AD { print\"un" "implemented\\n" "\"} sub gg{ $#"
"SS<0 and goto &__;push@SS,$SS[ $#SS]} sub uu{ "
"$#SS<1 and goto &__;$SS[ $#SS]+= $SS[$#SS-1];$SS"
"[$#SS-1]=$SS[$" "#SS]-$SS[$#SS-" "1]; $SS[$#SS]-"
"=$SS[$#SS-1]}");
}Compiling this you can generate:
#!/usr/bin/env perl
use warnings;_:$_= <> ; defined or exit; @ARGV=split; __:$_= shift;defined or goto _; chomp ;(m*^\x70oO\x74$*i)?(print "$_ w\x61s h\145r\x65!\n"):((m*^\s\*(-?\d+(\.\d+)?)\s\*$*)?(push@SS ,$1):(&{chr(((ord)%39)+3**4)x2 } )); goto __; sub ff { @SS= ()} sub __{print "stack empty\n"} sub ss{$#SS<0 and goto & __; print $SS[$#SS]."\n"} sub SS{ $#SS<0 and goto &__ ; print pop @SS}sub _ { print "divide by zero\n"}sub ii{map{ print"$_\n" } reverse @SS} sub AUTOLOAD { print"unimplemented\n"} sub gg{ $#SS<0 and goto &__;push@SS,$SS[ $#SS]} sub uu{ $#SS<1 and goto &__;$SS[ $#SS]+= $SS[$#SS-1];$SS[$#SS-1]=$SS[$#SS]-$SS[$#SS-1]; $SS[$#SS]-=$SS[$#SS-1]}Here is example input/output:
p
stack empty
5 p
5
5 + 5 p
unimplemented
5
P
5
p
5So you can see that if one uses operators it just shows ‘unimplemented’ (observe how this does not happen in the C program in linux!) but if one uses ‘p’ by itself it will print what’s on the stack (for example).
To not get ‘unimplemented’ you can remove from the perl:
sub AUTOLOAD { print"unimplemented\n"}(Note that the author stated this was used for exception handling so another possibility is that the C code doesn’t generate the correct perl and indeed even spaces is known to break the perl in at least one way, syntax errors, but maybe others too.)
Here is an example run of several operations in the perl script after removing that subroutine:
$ echo 5 5 p | ./dlowe.pl
5
$ echo 5 5 + p | ./dlowe.pl
5
Undefined subroutine &main::UU called at ./pl.pl line 2, <> line 1.
$ echo 5 5 * p | ./dlowe.pl
5
Undefined subroutine &main::ww called at ./pl.pl line 2, <> line 1.
$ echo 5 5 - p | ./dlowe.pl
5
Undefined subroutine &main::WW called at ./pl.pl line 2, <> line 1.
$ echo 5 5 / p | ./dlowe.pl
5
Undefined subroutine &main::YY called at ./pl.pl line 2, <> line 1.
$ echo 5 P | ./dlowe.pl
5$ echo p | ./dlowe.pl
stack empty
$ echo P | ./dlowe.pl
stack emptyNote the undefined subroutine error: that happens with commands that
are not even in the program. Again the author used AUTOLOAD
for exception handling so it seems likely that there is something
missing (which the below shows as well). For instance inputting
D followed by XX:
Undefined subroutine &main::nn called at ./pl.pl line 2, <> line 6.
Undefined subroutine &main::[[ called at ./pl.pl line 2, <> line 1.That means then that in linux despite the subroutines seemingly not existing it works in the code, at least in some cases.
As far as the unimplemented error goes: the author stated that it means there is invalid input. But as can be seen by the input I (Cody) gave the input is not actually incorrect. It is also curious to note that it ends up printing what’s on the stack at that point:
$ echo 5 5 + p | ./dlowe.pl
unimplemented
5which might (?) suggest that the + operator is
unimplemented. Unfortunately it has been many years since I have used
perl and I was never a guru either.
This program does not do what you might think it does! Running it like:
./primenum 13will seemingly wait for input exactly because it is waiting for input. See the README.md file or look at the source.
Although the name of the program suggests it prints prime numbers this is not the case. This is by design. See the author’s comments for more details or better yet look at the code and if necessary try it out. Please do not try and fix this.
A crash in the program is known as well. This is also a feature. Please do not try to fix the crashing of this code except to challenge yourself (if you think that it’ll be worth your two second fix :-) ). If you do fix it please do not make a pull request.
If DISPLAY is not set the program will very likely
crash, do something strange (or if you’re very unlucky your computer
might halt
and catch fire! :-) ).
Cody Boone Ferguson fixed this so that it doesn’t segfault and then also fixed the functionality of it (but see below).
Cody also fixed the anonymous.ten.c program (it
also segfaulted) but notes that it MUST be compiled as
a 32-bit ELF binary so it will NOT work in macOS (for
example); if you cannot use -m32 or even if you can but it
cannot compile as an ELF binary (not elf binary :-) ) (but see below)
then the anonymous program itself will very likely crash if
you run it on the compiled anonymous.ten.c but if nothing
else it will not modify the target executable (this part of the fix at
least should be correct).
The file 2001/anonymous/anonymous.ten.32 is a
pre-compiled 2001/anonymous.ten.c as
a 32-bit ELF binary in case you can run ELF binaries but cannot compile
32-bit binaries.
Other BSD Unices were not tested.
Note also that if you don’t specify a file or you specify a non-32-bit ELF file this program will very likely crash or do something strange like slaughter the elves of Imladris :-(
The two statuses might seem contradictory but that is a complicated question. The author stated that it only works with i386 linux so on the one hand the fact it doesn’t work in modern systems is considered a feature and not a bug. But on the other hand it would be nice if there was an alternate version which worked for modern systems. This does seem quite unlikely but some fixes, described next, were made.
Cody Boone Ferguson fixed an initial segfault (trying to open the file) and he also fixed the supplementary program bellard.otccex.c but this still crashes in systems not i386 linux which is what the author stated.
On the other hand if you do have a fix for 64-bit systems you’re welcome to provide it as an alternate version. If by chance you have a fix so that it works for both 32-bit and 64-bit systems that is also okay.
The author said that they compiled it with gcc version 2.95.2. We don’t know if a certain gcc version is necessary but it might be helpful to download and compile that version to test it - or it might not.
I (Cody) have no i386 system to test this but perhaps this is why I
can’t get it to work. Yusuke was able to get this to work with
-m32 but only in an emulator.
On the author’s web page for this program it is explicitly stated that it requires i386 linux.
There I found what should be a more portable version which is
included as otccelf.c (after adding
some #includes and the modification by Yusuke noted in the
thanks file) but it appears this
also requires i386 linux; indeed looking at the code it hard
codes paths that are i386 specific to linux.
Another point of interest is that the author provided de-obfuscated versions which might be of value to look at. I might do that as well but this entry is very likely never going to work for 64-bit linux so that’s not that likely since there are other things that are more important.
If you have a fix for 64-bit systems this is welcome as an alternate version, as stated above. You might like to look at the otccelf version but note that it (at least in 64-bit linux and macOS) has compilation errors.
This is a good question. The reason is we believe it better to fix some obvious problems: there were some bugs that would very possibly prevent it from working even if it was in i386 linux though Yusuke seemed to get it to work in an emulator so perhaps not. Still it’s better to have the type of the file pointer correct and so that the file can at least be opened in modern systems even if the compiler won’t work there.
Also the supplementary program, which did not work at all, was fixed (by Cody) and it can be run by itself for fun in modern systems, which was not possible before the fixes there.
This program deliberately crashes if it loses (which is what it aims to do).
The author referred to the file herrmann1.turing but it
does not exist not even in the archive. Do you have a copy? Please
provide it!
There is also a bug in part. During compilation you’re supposed to see some animation but this does not seem to work with modern gcc versions. It appears that version 2.95 works but maybe others do as well. Do you have a fix? We would appreciate your help!
If you want to try and fix this (mis)feature, you are welcome to try.
Sometimes when one player presses q it will result in
broken pipe on the other end.
The author also noted that some versions of curses have
problems with getch() having a delay.
The program is very likely to segfault if the specified port can’t be bound or connected to.
If the terminal is too wide (the author suggests over 500 characters) the program can segfault.
It’s also possible to crash the program if a player scores over a billion points.
Although it is independent of endianness both systems need the same character set. In other words both have to be ASCII or EBCDIC - not one of each.
The author stated that there is a cat man page for this program in case one wanted to install it as a tool but this is missing.
The glob pattern must match the whole string. See the author’s comments for details and a workaround.
There’s also no way to escape meta characters.
This function uses gets(3) but it would be ideal if it
used fgets(3). This one is rather complicated and will be
looked at later (some progress was made by Cody but it is less important
than other work) but you are welcome to try and fix this if you
wish.
This program only has one arg to main(). However
changing it to have 2 is not as simple as it might seem. Doing this
breaks things. If memory serves changing it to two will, for instance,
break the punch card code output. It might be that that is for
fgets(3), however, but the point remains: in order to test
any fixes you must determine if the extra files generated (some from
sort(1)) - see the README.md file - generate correct output
as well as that the main entry does too. You might try running the
try.sh script and redirect it to a file first so that you
can compare the output. Make sure to recreate the extra files as
described by the author if you do fix this. This might be looked at
later if nobody else takes up the challenge.
The author referred to a file card.gif but this file is
missing. Do you have it? Please provide it!
There seem to be a couple bugs at least in this entry. The first one
is that when it does reach a certain point it crashes. This is in the
function J() (or one it calls?). But sometimes it seems to
not get that far, just showing lines moving back and forth. This is also
in J(). It is not confirmed but it might be that the lines
are drawn by XDrawString(), the last part in
J().
J() has a for loop identified by
for and also a couple while loops, identified
by R.
Since the game never ends properly the score is not printed either.
It appears that the crash happens in more than one place. For instance it has happened at
XCheckMaskEvent(d, 4,&e)but it’s also happened at
else XDrawPoint(d ,w,g,(l.s+=l.a)>>9, h=(l.c+=l.b)>>9)In both cases the pointer d was corrupted:
(lldb) p d
(Display *) 0x0000000000000001It is believed that it also happened at
XDrawString(d,w,g,W/3,H/2,m,B);in J() and for the same reason.
The loop seems to be stuck in the call to usleep(3)
like:
usleep(p*200);which happens to be in a while() loop.
There is a way to get the code to crash. If you don’t do anything until everything is destroyed (where you’d lose the game if it ended) and then you start to shoot your missiles repeatedly it will eventually crash. It appears not to be the case (though this was not tried as many times) if you start shooting when the missiles come down. The fact you can shoot after it starts that sleeping in a loop does suggest that it’s not stuck only showing those lines and sleeping.
There was no IOCCC in 2002.
There was no IOCCC in 2003.
Segmentation fault will occur in some systems. For instance on macOS with the arm64 chip:
./gavin > kernelWhen trying to link gavin.o to produce sh,
the linker generates:
Undefined symbols for architecture arm64:
"start", referenced from:
-u command line option
(maybe you meant: __start)
ld: symbol(s) not found for architecture arm64We believe that this simply won’t work with non x86 specific systems but perhaps you have a fix? We welcome your help!
Although not related some recent changes were made to 2004/gavin to let it compile under clang. The following patch was applied:
diff --git a/2004/gavin/gavin.c b/2004/gavin/gavin.c
index c967b7e..2082b49 100644
--- a/2004/gavin/gavin.c
+++ b/2004/gavin/gavin.c
@@ -1,8 +1,9 @@
+int main(int t, char **q, char **d) { return cain(t, (int)q, (int)d); }
#define G(n) int n(int t, int q, int d)
#define X(p,t,s) (p>=t&&p<(t+s)&&(p-(t)&1023)<(s&1023))
#define U(m) *((signed char *)(m))
#define F if(!--q){
-#define I(s) (int)main-(int)s
+#define I(s) (int)cain-(int)s
#define P(s,c,k) for(h=0; h>>14==0; h+=129)Y(16*c+h/1024+Y(V+36))&128>>(h&7)?U(s+(h&15367))=k:k
G (B)
@@ -52,7 +53,7 @@ else
G (_);
G (o);
-G (main)
+G (cain)
{
Z, k = K;
if (!t)The original Makefile from 2004 had the following to say about this entry:
# Special flags for the gavin entry
#
# FYI: Older versions of GCC have a bug in -O2 optimization, hence -O1
#
GAVIN_OPT= -O1
GAVIN_FLAGS=\
'-DY(m)=*((int*)(m))'\
'-DE(f,a,b,c)=((G((*)))(f))(a,b,c)'\
'-DM=for(D=0;D<786432;D++)'\
'-DZ=int i=0,j=0,h,n,p=393728,s=26739,C,D'\
'-DV=0x90200'\
'-DK=0'\
'-DR=while((C=E(V-8,100,0,0))&3&&(D=E(V-8,96,0,0))|3){'\
'-DL(c,d)=E(V+8,100,c,0);R}E(V+8,96,d,0);R}'
# ...
# Best of Show
#
gavin: gavin.c
${CC} ${GAVIN_OPT} -o gavin ${GAVIN_FLAGS} gavin.c
${RM} -f kernel
./gavin > kernel
${CC} ${GAVIN_OPT} -c ${GAVIN_FLAGS} -DB=_start '-Dputchar(a)=' gavin.c
${RM} -f sh
${LD} -s -o sh gavin.o
${RM} -f vi
${CP} sh vi
${RM} -f fs.tar
${TAR} -cvf fs.tar sh vi gavin.c README.md prim
gavin_clean:
${RM} -f sh vi kernel gavin.o
gavin_clobber: gavin_clean
${RM} -f gavin fs.tar
gavin_files: boot.b lilo.conf prim gavin_install.txtThe current (Makefile was modified to try and fit into the current IOCCC build environment.
See known features in the README.md for things that are not bugs but documented (mis)features.
The author stated that:
Arguments matching `/[^2]22$/` cause the program to segfault
after rapidly exhausting the available stack space. I could
probably fix this by adding some special cases to the code,
but [Blaine the Mono] it's quite a bit more exciting this way,
don't you think?[/Blaine] Since the trailing `22` will be
its own atom anyway, you can just insert a space in front of
it if you run into this feature (e.g. `1 22` instead of `122`).
Arguments matching `/(?:.){10,}/` will produce visually incorrect
results (e.g., generation 1 starting from 1111111111 will
print as `: 1`), but there's nothing interesting about these
arguments that can't be modeled using shorter runs of the same
symbol.
Arguments matching `/(?:.){257,}/` may produce mathematically
incorrect results (e.g., generation 1 starting from a string
of 257 `1`s will be calculated as `11`); the remark from the
previous paragraph applies here also.The generated code will very likely segfault or do something not intended if not given the right args. See the README.md file for the correct syntax.
This program sometimes will create unsolvable puzzles :-) just to hook you. As a protection against this - and to prevent you from spending too much time on such puzzles (or too big puzzles if you insist on doing this) - it deliberately destroys its runtime stack. See the README.md for more details on this.
If you specify more than three args the program might also crash or
do something strange. This might also happen if you specify excessively
large board dimensions. Try 100 100 100 for instance and
see what happens!
This entry will very likely segfault or do something strange if the source code does not exist.
This entry requires that sed and make are
in the path.
The author states:
The only built-in functions are defvar,
format, substitute, char-upcase,
+, -, *, and
/.
Only one variable is available. This variable is called
q, although any symbol name starting with the letter
q will be mapped to this variable.
Only the first letter of a function name is significant. For
example, format, f, farm, and
various other words starting with f will all map to the
format function.
The defvar function completely ignores its first
parameter. It will always store the result of evaluating its second
parameter in q. Unlike in Common Lisp,
multiple invocations of defvar on the same variable will
overwrite old values.
Character constants are limited to #\Newline and
single-character constants such as #\x.
The char-upcase function will give strange results
if passed something other than a lowercase character.
The format function takes exactly three parameters.
The first parameter is ignored; output always goes to
stdout. The second parameter must have exactly one
~s format specifier in it, and no other format specifiers
are permitted.
The arithmetic functions take exactly two parameters.
All tokens must be separated from a following ) by
whitespace.
A lot of other things. Notably, the language which this program interprets is nowhere near Turing-complete.
Basically, the [LISP][] interpreter is good for some basic math operations, and for running itself.
Cody Boone Ferguson notes that, though probably obvious, this entry will not work with https. He added an alt version that scans for https in case someone wants to set up a pipeline or other workaround (perhaps with stunnel) or even add additional code to it to make it work with https but the problem otherwise is a secure connection has to first be set up in order to give http commands. This is not a bug but it’s worth pointing out as it won’t work on as many websites as it used to including the IOCCC website itself.
The author stated the below points of interest.
The machine emulated is an older 40 column one so 80 column programs will not show up correctly.
There is no emulation of PET graphics characters.
The parameter controls the speed of the 60Hz “jiffy clock”, and not the processor speed. This means many games may run too fast to be usable - it really depends on the speed of your machine.
Although the PET emulator can do
LOAD and SAVE, it cannot VERIFY
or OPEN and CLOSE files.
The PET hardware emulation is not at all complete - features such as the hardware timers are completely missing. Some programs will not run correctly.
The 6502 emulation does not include the seldom used decimal mode, or any of the “undocumented” instructions.
This entry uses gets(3) which is unsafe and provides
annoying warnings, most obnoxious in macOS as it shows it at runtime
(redirecting stderr to /dev/null will silence
it). The following diff will almost work but it crashes with at least
computer.tofu input file:
12a13
#define gets(c) fgets((c),PI,stdin)&&(((c)[strlen((c))-1]='\0'),c!=NULL)The string specified must be <= 42 characters and may only consist
of the characters in the regex a-z_A-Z0-9@.-. Breaking
these constraints will end up with possibly corrupt GIF files.
This program requires x86 (with an x87 FPU) or x86_64 machine and it requires the SDL1 library.
This program also requires being able to read, write and execute memory. With the Apple silicon chips this is not allowed. Fixing this might be quite challenging but you are welcome to try and fix it.
If you do fix this it might be good to fix the alternate code as well or else provide it as an additional alt version. Fixing this is very likely to be very challenging and in some systems it will not be possible to fix but you are welcome to try and fix it if you wish to!
Incorrect formulas will ungracefully crash the program.
The author stated:
If you pick a number higher than 19186, the program will return a
solution but it will be a rotation of one of the first 19186. This is
because the cross shaped piece fits 48 ways in the 4x4 cube, but only
2 of those ways are unique - you can rotate one of those to make any
of the other 46. The algorithm used always places the cross piece
first, so the first two placings of that piece result in the 19186
unique solutions.
If you pick a number higher than 460464 (24x19186) the program will
return without outputting a solution. If you can wait that long.Cody Boone Ferguson
fixed this program to not crash in macOS (it appeared to work fine in
fedora linux on an x86_64 architecture but he does not know
if it crashed in macOS Ventura due to the fact it is an
arm64 CPU - the Apple silicon chip). This fix, BTW, is a
mix of a fix by Yusuke Endoh
and Cody’s. Nevertheless there are two features that are not bugs. Later
the author made some slight changes to make it not crash in x86_64
macOS.
By design this program is supposed to crash on termination. This actually no longer seems to happen but we don’t need this to be added even if it might be called a bug :-)
You must type in caps (except in strings) and this program is indeed case-sensitive.
There was no IOCCC in 2007.
There was no IOCCC in 2008.
There was no IOCCC in 2009.
There was no IOCCC in 2010.
For a great amount of data points the program will crash, depending on your available stack space.
Giving too few (distinct) points can also cause a crash or division by zero.
The program expects plain numbers separated by whitespace.
This program assumes ASCII.
If you give the program any arg it might flood the screen.
It requires both syscalls fork(2) and
wait(2).
Giving any arg to the program is likely to flood your screen.
Rounding errors might cause an omission in the highest bin. There might be empty bins at the edges.
The author’s website, http://www.pootpoot.net, no longer exists as it once did, instead being something else entirely. The Internet Wayback Machine, although it archived it, did not load scripts. Do you know if the domain was moved? Do you have an archive or mirror? Please provide us it! Thank you.
The author states the following:
Bad input (e.g. nonexistent files, non-numeric number of iterations, etc.) tends to result in empty output.
Given exactly one corpus, the program will crash or produce garbage.
Leaks memory and file descriptors while processing files.
Will crash and die horribly if it runs out of memory.
If you want to try and fix this (mis)feature, you are welcome to try.
The author stated that there are a number of features and limitations. As the list is rather long see other features and limitations and remarks instead.
This does not appear to work with macOS, resulting in a segfault (and sometimes a bus error).
We’re not sure if this is to do with the Apple silicon chip or not but it seems like it might (see Apple resources below) be but it could also be a wider problem with macOS. It appears that Apple does not allow certain combinations of memory protections with the arm64 processor.
Cody Boone Ferguson tried some of the workarounds but at this time he could not get it to work. He offers some earlier debugging sessions below as well as a resource from the author as well as some resources on Apple’s website should anyone wish to take a crack at it. He might address it later but a starting point might be in richards.alt.c.
At first glance it appeared to be that it might be the function
pointers or the fact it is trying to execute code in memory (as noted
above). The function pointers were changed a bit but this has not helped
solve the problem with macOS. A few things that were noticed depending
on asan sanitisers specified. With address in
linux:
richards.c:101:9: runtime error: null pointer passed as argument 2, which is declared to never be nullbut it actually works. This code is:
} else
p++;
W(c)} /* LINE 101 */
}Okay but what about argument 2? Well W is defined
as:
#define W(c) e=memmove(e,x[(u)c], y[(u)c])+y[(u)c];so it would appear that
x[(u)c](at least in my tired head?) is NULL. But why does it work then?
NOTE: u is int.
Under macOS (with the arm64 chip) we get:
UndefinedBehaviorSanitizer:DEADLYSIGNAL
==10834==ERROR: UndefinedBehaviorSanitizer: SEGV on unknown address 0xffffffffffffffff (pc 0x00019f446748 bp 0x00016f353100 sp 0x00016f352b90 T139066)
==10834==The signal is caused by a WRITE memory access.
#0 0x19f446748 in _platform_memmove+0xa8 (libsystem_platform.dylib:arm64e+0x3748) (BuildId: 756cd10d62a032839e57cbaa810c95ac32000000200000000100000000030d00)
#1 0x100c33c2c in main richards.c:202
#2 0x19f0bff24 (<unknown module>)
==10834==Register values:
x[0] = 0xffffffffffffffff x[1] = 0x0000000100ab18c1 x[2] = 0x00000000000000cf x[3] = 0x0000000000000000
x[4] = 0x0000000100ab18a1 x[5] = 0x0000000000000001 x[6] = 0x000000000000000a x[7] = 0x0000000000000000
x[8] = 0x00000001015972b8 x[9] = 0xffffffffffffffff x[10] = 0x0000000000000001 x[11] = 0x00000000000001f0
x[12] = 0x0000000000000001 x[13] = 0x00000000000003e0 x[14] = 0x0000000000000001 x[15] = 0x0000000000000084
x[16] = 0x000000019f4466a0 x[17] = 0x00000001ff3ad908 x[18] = 0x0000000000000000 x[19] = 0x0000000100c32a80
x[20] = 0x00000001017e0000 x[21] = 0x00000001017e1910 x[22] = 0x000000016f353530 x[23] = 0x000000019f13a366
x[24] = 0x000000016f3534b0 x[25] = 0x0000000000000001 x[26] = 0x0000000000000000 x[27] = 0x0000000000000000
x[28] = 0x0000000000000000 fp = 0x000000016f353100 lr = 0x0000000100aaf880 sp = 0x000000016f352b90
UndefinedBehaviorSanitizer can not provide additional info.
SUMMARY: UndefinedBehaviorSanitizer: SEGV (libsystem_platform.dylib:arm64e+0x3748) (BuildId: 756cd10d62a032839e57cbaa810c95ac32000000200000000100000000030d00) in _platform_memmove+0xa8
==10834==ABORTINGand naturally it does not work.
Adding the proper PROT_WRITE constant to the
mmap() call does not seem to help. Of course with the
address this is not surprising so why is a different address allocated
in macOS? Or is it?
Now with address sanitiser both linux and macOS crash
at:
=================================================================
==10983==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x00016f8d5e78 at pc 0x0001005dd910 bp 0x00016f8d5cd0 sp 0x00016f8d5cc8
READ of size 8 at 0x00016f8d5e78 thread T0
#0 0x1005dd90c in rd richards.c:132
#1 0x1005ddf8c in re richards.c:149
#2 0x1005110a0 in t richards.c:108
#3 0x1005df72c in main richards.c:189
#4 0x19f0bff24 (<unknown module>)
Address 0x00016f8d5e78 is located in stack of thread T0 at offset 56 in frame
#0 0x1005dd678 in rd richards.c:129
This frame has 2 object(s):
[32, 36) 'v.addr'
[48, 56) 'a' (line 130) <== Memory access at offset 56 overflows this variable
HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork
(longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: stack-buffer-overflow richards.c:132 in rd
Shadow bytes around the buggy address:
0x00702df3ab70: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3ab80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3ab90: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3aba0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3abb0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x00702df3abc0: 00 00 00 00 00 00 00 00 f1 f1 f1 f1 04 f2 00[f3]
0x00702df3abd0: f3 f3 f3 f3 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3abe0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3abf0: 00 00 00 00 f1 f1 f1 f1 00 f3 f3 f3 00 00 00 00
0x00702df3ac00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x00702df3ac10: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Shadow byte legend (one shadow byte represents 8 application bytes):
Addressable: 00
Partially addressable: 01 02 03 04 05 06 07
Heap left redzone: fa
Freed heap region: fd
Stack left redzone: f1
Stack mid redzone: f2
Stack right redzone: f3
Stack after return: f5
Stack use after scope: f8
Global redzone: f9
Global init order: f6
Poisoned by user: f7
Container overflow: fc
Array cookie: ac
Intra object redzone: bb
ASan internal: fe
Left alloca redzone: ca
Right alloca redzone: cb
==10983==ABORTING
/bin/bash: line 1: 10983 Abort trap: 6 ./richardswhich corresponds to the code:
o rd(u v)
{
o *a = 0, **b = &a, **c = b + 32;
x[v] = w;
for (; b < c && (*b < (o *)w || *b > (o *) T || *b == &v); b++); /* LINE 132 */but this seems to not necessarily be relevant since without it it works in linux.
A warning of interest when compiling is:
richards.c:169:15: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'o ()' (aka 'void ()') [-Wincompatible-pointer-types]
f *lib1[] = { T, T, T, T, T } ;
...
richards.c:170:15: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'o ()' (aka 'void ()') [-Wincompatible-pointer-types]
f *lib2[] = { T, ld, lp, lx, l1 } ;
^
richards.c:170:18: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'void (volatile dt)' (aka 'void (struct (unnamed at richards.c:36:3) *volatile)') [-Wincompatible-pointer-types]
f *lib2[] = { T, ld, lp, lx, l1 } ;
^~
richards.c:170:22: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'void (volatile dt)' (aka 'void (struct (unnamed at richards.c:36:3) *volatile)') [-Wincompatible-pointer-types]
f *lib2[] = { T, ld, lp, lx, l1 } ;
^~
richards.c:170:26: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'void (volatile dt)' (aka 'void (struct (unnamed at richards.c:36:3) *volatile)') [-Wincompatible-pointer-types]
f *lib2[] = { T, ld, lp, lx, l1 } ;
^~
richards.c:170:30: warning: incompatible pointer types initializing 'f *' (aka 'void (**)(volatile void *)') with an expression of type 'void (volatile dt)' (aka 'void (struct (unnamed at richards.c:36:3) *volatile)') [-Wincompatible-pointer-types]
f *lib2[] = { T, ld, lp, lx, l1 } ;
^~This can be fixed easily enough however but it doesn’t appear to matter in this case.
It might be helpful to use the try.alt.sh script to test that it does not crash and functions properly.
The author has more about the entry at https://github.com/GregorR/ioccc2011.
Porting Just-In-Time Compilers to Apple Silicon
Allow Execution of JIT-compiled Code Entitlement
Looking at https://github.com/apple/darwin-xnu/blob/5394bb038891708cd4ba748da79b90a33b19f82e/bsd/kern/kern_mman.c
Cody noticed a curious thing in the mprotect() function:
#if CONFIG_DYNAMIC_CODE_SIGNING and the protection variable
has the VM_PROT_TRUSTED set it might allow overriding the
problem but this is unconfirmed and it’s not known when
CONFIG_DYNAMIC_CODE_SIGNING would be defined.
Do you have a fix? We welcome it!
The author stated that the program will crash if no argument is passed to the program though we note that your computer might also halt and catch fire :-)
The author stated:
If there is a division by zero, square-root of a negative number, or similar
operation, then the results are undefined.The author stated:
The memory locations of `argv` are used for various purposes. Therefore, the
program name (default `deckmyn`) must be *at least 4 characters long*, including
possibly the path. This, along with the ending `'\0'`, gives a minimum 5 bytes of
useful memory space. Otherwise, the program may attempt to write outside the
string.
The code is quite sensitive. Errors in the input can lead to strange results.
The code does not read beyond the end of the music input, but that is about the
only error checking available. Any input that is not according to the rules in
the deckmyn.md, may cause errors.
The program has no special hardware limitations or requirements, other than 8bit
`char` (`signed` or `unsigned`) and two's complement negatives.
The limitations of using `char` as counters for e.g. the number of music staves
are minor. 127 staves to a page is rather a lot.
The file layout must be exactly the same as deckmyn.c though, only the locations
of space (`" "`) may differ.
Every element in the music is given by a string of *exactly* 3 characters. Any
violation of this will result in the rest of the music being interpreted in some
randomly wrong way. Notice that in a file, the newline is also a character!
Therefore, when preparing music in a file, care should be taken that the last
musical element of a line is only 2 characters!The manual referred to is here.
The author stated:
splash.d, dead.d and
sprites.d).The author stated:
If the input doesn't start with a valid program, that is,
if the interpreter reaches end-of-file during program parsing,
it will crash in some way. E.g. the following might dump core:
echo -n "U" | ./tromp
Furthermore, the interpreter requires the initial encoded lambda term to be closed,
that is, variable n can only appear within at least 'n' enclosing lambdas.
For instance, here the term '\ 5' is not closed, causing the interpreter to crash when
looking into a null-pointer environment:
echo ">Hello, world" | ./tromp
will likely dump core.
The author stated that the program will crash if no argument is passed to the program or if invalid arguments (e.g. file does not exist) or images of mismatching sizes or unsupported pixel formats though we note that your computer might also halt and catch fire :-)
The author stated:
This program will possibly crash or draw something strange with 0 args. Then again it might not. :-) This is easy to fix but would add bytes and since the author noted it is a feature and not a bug. You of course can try and fix it yourself but please do NOT open a pull request for this: just do it as an exercise to see if you understand the code.
You can try and answer the questions, too: when will it crash? When will it draw something funny (or will it? :-) ) and when will it just do nothing?
The author also stated:
From the author:
–
This program supports only “Combinator-calculus style notation” of Lazy K. “Unlambda style” and “Iota and Jot” style are not supported.
Also, it requires a space between identifiers. In short, use
(S K) instead of (SK), “`sk”,
**i*i*i*ii*i*i*ii, or 11111100011100.
Huge memory may be required to compile the program (about 300 MB on my machine).
In addition, there are some limitations (and workarounds) mentioned in Spoiler section.
–
From the author:
–
You can NOT write a note length immediately followed by a
note E, such as C2E2.
Can you figure out why?
A workaround is inserting a whitespace: C2 E2.
–
Invalid input files will very likely crash the program.
This program will not terminate on its own; you must kill
hou (but not Qiming Hou :-) ) yourself. This should not be
fixed.
The author reminds us that if you kill the program you will have to wait a short bit of time before you can start it again if you want it to be able to bind to the socket (which obviously you do). The source is out the scope of this document but the author does mention the well known (at least to those of us who have experience with socket programming :-) ) fix. However as the author pointed it out as a known limitation it is not a bug but a feature.
The author noted that in macOS the colours might be wrong. They gave a solution. However Cody didn’t see any problem with it and he beat the series many times. Of course the game here is small and he’s practically blind and it’s been a while since he last played so it might be wrong. Nevertheless the author does give a way to reconfigure the colours.
The author noted that the character cannot go through walls and the impression is that this is how it was believed but Cody ironically pointed out that there are known glitches where in some places you actually could go through walls. This might even have applied to the arcade version, Mario Bros, though that can no longer be confirmed by Cody. The point here is that in this game you cannot go through walls but it’s not a bug.
On the other hand, the author also stated:
–
When the Super character becomes bigger
(ZOOM flag), the character can collide with blocks and get
stuck inside them. This is a KNOWN BUG. When your player become bigger,
stay away from blocks!
–
but since it’s documented it’s considered a feature, not a bug to fix.
This program will very likely crash if no arg is given.
Cody Boone Ferguson
discovered a bug to do with translating sound to text that shows itself
in some cases (it works in others) when working on his winning 2020 Enigma machine (‘Most enigmatic’)
entry. See his README.md for
details (search for vik in the file). This one might be
more of a limitation but this is not known.
He provides a tip in testing the problem: it might help to use his Enigma machine to find problems as it will allow you to verify what is correct and what is not. Again see his README.md for details. However it is also possible to not rely on the entry as the below shows. One should be able to do:
echo 'No. I want chocolate!' | ./prog > chocolate.raw
./prog e < chocolate.rawand get:
No. I want some chocolate!but instead we get:
IT FHOCOLITE!which is very wrong. If however one does:
echo IOCCC | ./prog > ioccc.raw
./prog e < ioccc.rawthey will properly get:
IOCCCThe author stated:
–
Hard requirements:
The platform must implement the double type as
IEEE754-compliant 64-bit floating point numbers. The 80-bit intermediate
format used by x87 is considered as an violation of this. The code
should print an error message on such platforms.
The program must start with the CPU / FPU in round-to-nearest mode.
Soft requirements:
The compiler must respect volatile. The code is
formatted to warn about that, though.
The printed result is only correct on little-endian machines. The program takes care to warn about this issue after printing an incorrect big-endian result. Error messages become garbled, though.
–
The program doesn’t look at the header of files so if it’s passed something hat is not compressed data it’s likely to crash.
The program depends on little endian systems.
The program assumes that EOF is -1. This
can be fixed but at this time it is uncertain if it should be.
The author also stated:
–
While the program works best when bytes/characters are octets and the
number of bits in a type is sizeof(typ) << 3, it will
work correctly on 24-bit or 36-bit systems with 9 bits/byte, or systems
where sizeof(typ) is 1 even for int and so on.
On such systems, it will only use 8 * sizeof(typ) bits per
place. It does not work when CHAR_BIT <= 7.
–
There was no IOCCC in 2016.
There was no IOCCC in 2017.
The author wrote:
int is less than
32 bits.-d or -h is
fully possible if it is not the first file provided, even though I am
not sure why you would want to do that.When you run it on a JSON file you will see something like:
$ ./prog < ioccc.json > ioccc.html
Assertion failed: (rulez), function C, file prog.c, line 124.
Abort trap: 6(from say running
./try.sh) but this is expected and the fileioccc.html` will
be generated properly.
Some JSON files might cause the program to continue to run and
seemingly infinitely increase the size of the output file. This can
happen if you try fixing the syntax error in the generated
ioccc.json file.
After exiting the program (with ctrl-e) if you try running it again you will likely see:
>bootT5pw
]h_ 8D3[]where [] is the cursor. When this happens if you hit
enter (this is necessary or else it’ll happen again) and then exit again
(ctrl-e) and run prog again it’ll be okay.
The author stated that if you make a typo it can happen that the boot loader can crash and halt. If this is the case type ctrl-e to quit the emulator and try again.
The author also stated that it is possible to corrupt the virtual
disk. If you do this you should type make clobber all and
try again.
Cody caused a funny problem where it showed ERROR after
each command. The solution was to hit ctrl-e and start again. He does
not know if this is what was referred to by the author but there was
likely no typo involved.
Cody notes that if you are using the saved mode you must type
sync prior to exiting the program or else the next time you
run it the file will not exist (or in the case of compiled code it won’t
be executable).
The author wrote the following:
Despite appearances, it does not handle numbers in hex, or
provide a curses(3)-based interface.
The expected input format is zero or more lines of space-separated integers. If other characters appear in the input, it will either reject the input entirely and exit with a non-zero status, or skip number(s) adjacent to the non-digit characters, depending on where the characters appear. Tabs and multiple consecutive spaces are handled correctly, however.
Individual lines of input longer than 0x3543 - 1
bytes will be split and processed as if they were multiple lines of
input, which can produce incorrect results. This magic number’s
significance is described earlier.
The algorithm expects its input to represent a fully connected graph. While the output is otherwise topologically sorted, if there are nodes completely unconnected to the rest of the graph (with or without self-cycles), they will be output as soon as they are processed – this means that, when there are disconnected nodes, reordering the input lines can produce different output. Addressing this by adding another pass is would put the program over the size limit.
While the node IDs don’t need to be consecutive or start at 0, the implementation doesn’t have special handling for sparse graphs. If you give it a graph with nodes numbered 0 and 2147483647, it will attempt to allocate sufficient memory (potentially around 32 GB) for the entire range of graph nodes, even if those are the only ones. If memory allocation returns NULL, it will gracefully exit, otherwise it will succeed, eventually, perhaps after a great deal of swapping.
Node IDs >= 2147483648 will cause the program to print an error message and exit with a non-zero status. This shrinks the code that detects overflowing the array size by a bit.
A very large group of nodes in a cycle can cause a stack overflow. This typically takes over 100,000 nodes, and depends on the order the nodes are visited. Addressing this would put the program over the size limit.
The implementation depends on the characters '0',
'1', … '9' having the values 48
through 57, rather than using isdigit(3). As
noted above, this program has nothing to do with a hand.
Certain input can crash this program. The file crash.unl is an example file.
The author pointed out that some implementations of
wc(1) show different values but his implementation matches
that of macOS and FreeBSD.
Cody Boone Ferguson fixed the scripts so that they can work but it might end up that the scripts don’t show any output anyway as they use the full alphabet. To test that it works one can have a script like:
LC_ALL=C aäbcdefghijklmnoöpqrsßtuüvwxyz | LC_ALL=C grep .. | LC_ALL=de_DE.UTF-8 ./prog aäbcdefghijklmnoöpqrsßtuüvwxyzIf you run this you should see:
aäbcdefghijklmnoöpqrsßtuüvwxyzwhich shows it works. However as the de.sh script refers to all letters it can’t find a perfect pangram. BTW, as far as the question of whether the umlaut letters or the Eszett are considered letters in the German alphabet, Cody noted in the de.sh/de.alt.sh scripts:
# Are the umlauts ä, ö, ü and Eszett (ß) letters in the German alphabet? There
# is more than one opinion on the subject but a lot of words do have an umlaut
# (or Umlaut in German). In some (probably all) systems this script outputs
# nothing probably because it's very hard to form a perfect pangram in German,
# if it's not impossible. Nevertheless, we do include the other characters
# whether or not they are in your view considered part of the alphabet for the
# reason that so many words have them.The author does not it uses gets() and one will get
warnings but it would be ideal if this was not the case.
There are two things to be aware of with this entry.
If the file being used (to draw) is deleted it might lock any session still in use. These will have to be killed from another shell session or by closing the terminal tab. This will also happen if the file cannot be opened in the beginning.
This program will also very likely leave sockets lying about in the current working directory. For instance Cody Boone Ferguson showed us this:
$ ls -al |grep '^s'
srwxr-xr-x 1 cody staff 0 Apr 6 08:19 .BDHFHALG=
srwxr-xr-x 1 cody staff 0 Apr 6 08:15 .CGGHAMGC=
srwxr-xr-x 1 cody staff 0 Apr 6 08:16 .CMDGAELH=
srwxr-xr-x 1 cody staff 0 Apr 3 08:47 .CMLBCCDA=
[...]This is NOT a bug and you’ll have to (at least at this time?) delete the files manually. You shouldn’t have to worry about these being added to git: it seems to ignore sockets (it did at least in macOS).
He provides the following tips on this situation. A simpler way to find sockets in the directory:
file .*|grep socket|cut -f1 -d:To delete them you can do:
find . -exec file '{}' \;|grep socket|cut -f 1 -d: | xargs rm -fthough one might want to check that the program is not currently running. :-)
As a backtrace quine this entry is SUPPOSED to segfault so this should not be touched either.
The author stated the following:
The author wrote that there are a number of differences from what one might expect. Rather than duplicate the information we refer you to the author’s remarks in the sections Syntax and Caveats.
The author wrote that if you decide to change networks or use a different input file you should delete all the checkpoint files because the format depends on both the network and the input – using the wrong checkpoint is likely to cause a crash.
The author also wrote:
IMPORTANT NOTE: Since OMLET uses the system stack
for network storage, larger networks may cause OMLET to crash (typically
with a message like Segmentation fault) unless the system
stack size is first increased. The exact command for doing so depends on
your shell and your system’s hard limits. On
sh/ksh/bash shells, you can view
the hard limit with ulimit -Hs and set it with
ulimit -s 65532 (replacing 65532 with the
actual hard limit). On csh/tcsh shells, you
can view the hard limit with limit -h stacksize and set it
with limit stacksize 65532 (replacing 65532
with the actual hard limit).
This program will not validate input so it might fail or get stuck if invoked erroneously.
Also, the maximum file size is 1GB.
The author noted that if the program runs out of memory it is likely to crash.
The author also stated that ‘if input contains CR-LF end of line
sequences, those CRs are silently dropped. In fact, most
control codes are silently ignored except line feeds (preserved) and
tabs (expanded to 8 spaces).’
This entry is known to crash if no arg is specified. Although easy to fix it is documented and should not be fixed (of course you may fix it to see if you can but it shouldn’t be made into a pull request).
It will also show funny output with more than one arg. This should not be fixed either. But can you figure out why this happens?
The author stated that bad things happen if the entered move is
outside of the range [1..9]. We observed a crash but we also observed a
tie and other things besides (say with the input -1,
-, -10 and others).
If a player makes a move over their opponent’s place they will automatically lose.
If a player tries to play a move they already played they will forfeit their move.
If you have audible bells enabled the program will beep at every turn.
There are some things that might appear to be bugs but are actually features or things that are misinterpreted as bugs. See the bugs.md and troubleshooting.md files for details.
The author noted that the program only supports WAV files that have exactly 16 bits per sample, but it allows any sample rate and any number of audio channels.
The author listed the following limitations:
The program does not support MIDI files over 4 MB (specifically,
it will not take input files larger than 2^22 - 1 bytes).
This is unlikely to be a big limitation in practice, since even lengthy
MIDIs are rarely over a couple hundred KB.
Not all MIDIs set (or correctly set) their timing information, which will lead to the tempo of the drum beat not matching the track. This will be especially true for “live performance” .mid files.
There may be some types of MIDI messages that the program is unable to correctly parse, but I have rarely seen this in practice. The largest cause of program failures, in my experience, has been Limitation 2.
See also the Program error codes written by the author which lists some other conditions which should be considered features, not bugs.
There was no IOCCC in 2021.
There was no IOCCC in 2022.
There was no IOCCC in 2023.
We hope to run the IOCCCMOCK this year, 2024, and hopefully the next IOCCC, but for now we wish everyone a happy new year!
We hope this document was of use to you in determining which entries are known to have a problem, what entries are known to have features that are not bugs and so on. We also thank you for going through this document and, if you propose any fixes] via a GitHub pull request or otherwise, we thank you as well for the help! We will happily add you to the thanks file as well.