fix topbar to work on mobile devices

The topbar, by way of JavaScript, now works
on mobile devices such as cell phones and
tablets.

Internal links for entries from 1984 thru 2015
have been adjusted.

Ran `make www` to perform the above.

TODO: Fix some color issues on topbar.

TODO: More testing!
This commit is contained in:
Landon Curt Noll committed 2024-04-22 15:05:59 -07:00
1 parent 99cd17401e
commit 97f6da1247
440 files changed
+74125 -6956

No files matched your search

+15 -15
View File
@@ -65,7 +65,7 @@ program.
This program will loop infinitely while progressively refining a
[raytraced](https://en.wikipedia.org/wiki/Ray_tracing_(graphics)) image.
NOTE: the author refers to [a.c](a.c), placed in a gzipped file `a.c.gz`. We do not
NOTE: the author refers to [a.c](%%REPO_URL%%/2013/hou/a.c), placed in a gzipped file `a.c.gz`. We do not
include it but it can be generated like:
``` <!---sh-->
@@ -134,7 +134,7 @@ don't have to worry about conflicts.
### Abuse of the rules
* [hou.c](hou.c) uses compression to get around the size limit. The compression
* [hou.c](%%REPO_URL%%/2013/hou/hou.c) uses compression to get around the size limit. The compression
fully complies with the rules and the guidelines (at least the portion that
shows up in `grep` "size limit"). Please see the [Self-imposed
restrictions](self-imposed-restrictions) section below for more details.
@@ -145,18 +145,18 @@ restrictions](self-imposed-restrictions) section below for more details.
* The building process does not involve any OS tool beyond `cc(1)` and
`make(1)`. No `gzip(1)` compression! `a.c.gz` doesn't fit in 2053 bytes, anyway.
* Neither [hou.c](hou.c) nor [a.c](a.c) (the *real* decompressed source) uses
* Neither [hou.c](%%REPO_URL%%/2013/hou/hou.c) nor [a.c](%%REPO_URL%%/2013/hou/a.c) (the *real* decompressed source) uses
`#define` (or `cc -D`) at all.
* The source code is not required at runtime.
* [a.c](a.c) does not drop optional features to reduce size. There are pure
* [a.c](%%REPO_URL%%/2013/hou/a.c) does not drop optional features to reduce size. There are pure
optimization code that can be dropped without affecting the converged output
(only affecting the ray tracing speed / convergence rate). All files are
properly `fopen()`ed with `"rb"` / `"wb"` for Windows compatibility. The PPM
header has a comment line for non-standard-compliant viewers (specifically, my
old HDRShop 1.0). And there is a nice text message saying "please wait...".
Despite the messy look, [a.c](a.c) and [hou.c](hou.c) compile warning-free
([hou.c](hou.c) even wastes 18 bytes on `#include<stdio.h>` just for `putchar(3)`).
[a.c](a.c) compiles mostly clean in the C99/ANSI modes of clang and gcc (with:
Despite the messy look, [a.c](%%REPO_URL%%/2013/hou/a.c) and [hou.c](%%REPO_URL%%/2013/hou/hou.c) compile warning-free
([hou.c](%%REPO_URL%%/2013/hou/hou.c) even wastes 18 bytes on `#include<stdio.h>` just for `putchar(3)`).
[a.c](%%REPO_URL%%/2013/hou/a.c) compiles mostly clean in the C99/ANSI modes of clang and gcc (with:
```
-Wall --pedantic
@@ -178,15 +178,15 @@ internal states.
the code, but they leave the text clear in the *result*. This entry takes it
further and obfuscates the output image as well. Can you find the text in the
image? Hint: look up.
* [a.c](a.c) leaves all shaders in plain text, but the plain text shader code can't
* [a.c](%%REPO_URL%%/2013/hou/a.c) leaves all shaders in plain text, but the plain text shader code can't
be taken for its face value; the arithmetic rules subtly diverge from our common
sense.
* [a.c](a.c) is less portable than [hou.c](hou.c) itself. [hou.c](hou.c) only depends
* [a.c](%%REPO_URL%%/2013/hou/a.c) is less portable than [hou.c](%%REPO_URL%%/2013/hou/hou.c) itself. [hou.c](%%REPO_URL%%/2013/hou/hou.c) only depends
on ASCII and should run just fine on 16-bit, small memory, or
floating-point-incapable machines. [a.c](a.c), while still reasonably portable, is
floating-point-incapable machines. [a.c](%%REPO_URL%%/2013/hou/a.c), while still reasonably portable, is
quite memory consuming, requires IEEE754-compliant `double`, and assumes `int` to be
32-bits.
* Though technically endian-dependent, [a.c](a.c) remains portable providing that one
* Though technically endian-dependent, [a.c](%%REPO_URL%%/2013/hou/a.c) remains portable providing that one
doesn't copy saved sessions across different endianness.
@@ -216,7 +216,7 @@ corresponds to the final image, which can then be produced as a simple per-pixel
histogram of all generated paths. The robustness comes from a state mutation
strategy that actively tries to explore the neighborhood of high energy peaks
(e.g. paths that happen to hit the light source in
[otherroom.scene](otherroom.scene)). In addition, a rudimentary form of lens
[otherroom.scene](%%REPO_URL%%/2013/hou/otherroom.scene)). In addition, a rudimentary form of lens
path stratification is added to balance the attention each pixel receives. The
Metropolis-Hasting process completely avoids the tell-tale pixel sampling loop
required in most other image generation methods.
@@ -237,11 +237,11 @@ necessarily map to the same set of paths in `PSSMLT`. After all,
Metropolis-Hasting used an even worse PRNG in their 1953 paper.
The `PPM` compressor uses statically weighted fixed order contexts with an
arithmetic encoder tweaked for [iocccsize.c](../iocccsize.c). The encoder emits octet-space pairs
arithmetic encoder tweaked for [iocccsize.c](%%REPO_URL%%/2013/iocccsize.c). The encoder emits octet-space pairs
where each octet encodes ~6.5 bits of information and each space encodes 2 bits
(thanks to the generous definition of "space" in [iocccsize.c](../iocccsize.c)). The compressor
(thanks to the generous definition of "space" in [iocccsize.c](%%REPO_URL%%/2013/iocccsize.c)). The compressor
actively shuffles the variable names around until the compressed string happens
to contain enough `{}; `s to pass the final [iocccsize.c](../iocccsize.c) test.
to contain enough `{}; `s to pass the final [iocccsize.c](%%REPO_URL%%/2013/iocccsize.c) test.
There are a few other tweaks:
* The `PPM` model uses mostly whitespace characters for weights.