| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |\ \ \ |
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Arrays are just a better idea for storing multiple strings than relying
on word splitting. Consequently, several global variables in libs/*
were switched to arrays and any references to said variables modified
to expand to the arrays' elements.
|
| |\ \ \ \ |
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
It turns out that we only need bare HTML files for .md files in news,
yet publish.sh creates them unconditionally. That is, we spend a lot
of time building bare HTML files that we never use.
This commit makes it so that bare HTML files are only generated for
news files, which speeds up the website build significantly:
$ /home/michael/benchmark.sh speed-up-build master
Already on 'speed-up-build'
NOW TESTING ON speed-up-build
0m08.24s real 0m08.53s user 0m05.57s system
0m08.21s real 0m08.39s user 0m05.58s system
0m08.26s real 0m08.23s user 0m05.70s system
0m08.26s real 0m08.27s user 0m05.91s system
0m08.24s real 0m08.36s user 0m05.63s system
0m08.28s real 0m08.40s user 0m05.67s system
0m08.29s real 0m08.21s user 0m05.83s system
0m08.23s real 0m08.12s user 0m05.80s system
0m08.32s real 0m08.32s user 0m05.75s system
0m08.30s real 0m08.40s user 0m05.61s system
Switched to branch 'master'
Your branch is ahead of 'origin/master' by 33 commits.
NOW TESTING ON master
0m12.98s real 0m15.07s user 0m07.18s system
0m12.93s real 0m14.57s user 0m07.69s system
0m12.98s real 0m15.06s user 0m07.46s system
0m12.98s real 0m14.75s user 0m07.67s system
0m12.94s real 0m15.10s user 0m07.22s system
0m12.94s real 0m14.95s user 0m07.22s system
0m12.98s real 0m14.57s user 0m08.02s system
0m12.96s real 0m14.84s user 0m07.41s system
0m12.96s real 0m14.99s user 0m07.49s system
0m13.06s real 0m14.91s user 0m07.54s system
And here's the script in question, benchmark.sh:
#!/bin/sh
set -u
set -e
# usage: runit branch
runit() {
git checkout "$1"
echo
echo NOW TESTING ON "$1"
echo
for i in `jot 10`; do make clean >/dev/null; time make -j2 >/dev/null; done
}
make clean >/dev/null
for branch in "$@"
do
runit "$branch"
done
|
| |/ / / /
| | | |
| | | |
| | | |
| | | |
| | | | |
Unlike what the comment says, the target .md.html does apply to
news/index.md (it's just that news/index.md must be generated first
by index.sh).
|
| |\ \ \ \
| |_|/ /
|/| | | |
|
| |/ / /
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
When running "bash -x publish.sh index.md", the following happens:
+ pandoc --smart /tmp/libreboot_www.Xne7SYgf4e -s --css /global.css '--css /headercenter.css' --template template.html --metadata return=
pandoc: unrecognized option `--css /headercenter.css'
Try pandoc --help for more information.
This is because the OPTS variable is being quoted, which causes its
value of "-css /headercenter.css" to not be split on word boundaries,
that is, spaces. Because pandoc has a "--css" option, but not a
"--css /headercenter.css" option, pandoc expectedly complains that
said option does not exist.
To fix this we just unquote OPTS, and add a shellcheck ignore directive
so this won't be reintroduced in the future.
Caused by: a8d89665de6da20a8793886e03f153e922f6e519.
|
| |\ \ \ |
|
| | |/ /
| | |
| | |
| | |
| | |
| | |
| | | |
All of the warnings were of type SC2086 [1]: Double quote to prevent
globbing and word splitting.
[1]: https://github.com/koalaman/shellcheck/wiki/Sc2086
|
| |\ \ \
| |/ /
|/| | |
|
| | | |
| | |
| | |
| | |
| | |
| | | |
Splitting rss() into 3 smaller functions allows factoring the for-loop
out of rss() This is done by having the new rss_main() function take an
argument instead of operating on a global variable, as rss() did before.
|
| | | |
| | |
| | |
| | |
| | | |
I find it easier to read this file if all function definitions come
first, then all the actual stuff we do comes at the end.
|
| | | |
| | |
| | |
| | |
| | | |
It makes the code easier to understand than the global variables that
were used before.
|
| |/ / |
|
| | | |
|
| |\ \ |
|
| |/ /
| |
| |
| |
| |
| |
| | |
Replace brace expansions with extended globs in a couple of places where
brace expansions were erroneously used in place of actual pattern
matching. This avoids potential errors concerning nonexisting files
when patching sources.
|
| | |
| |
| |
| | |
no need for pull request on this one.
|
| |\ \
| | |
| | |
| | | |
master
|
| |/ / |
|
| |\ \ |
|
| |/ / |
|
| |\ \ |
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When running index.sh on OpenBSD, the following error happens for
each item in the news/ directory (output is from "bash -x"):
+ touch -d '4 Jun 2017' news/andrew-robbins-new-maintainer.md
touch: out of range or illegal time specification: YYYY-MM-DDThh:mm:ss[.frac][Z]
This is because OpenBSD's touch(1) requires that the "d" flag's argument
be in ISO 8601 format, that is, "YYYY-MM-DDThh:mm:ss[.frac][Z]". This
could have been dealt with by converting the article date (determined
by "sed -n 3p $f | sed -e 's/^..//g'") to ISO 8601 format, then passing
the date to touch(1). That would have required even more code, so was
discarded as a possible solution.
Instead, we solve this by keeping a MANIFEST file under news/, which
is read to determine (a) which articles should be added to news/index.md,
and (b) in which order. This avoids the need for touch(1) altogether,
finally making the whole libreboot website build properly on OpenBSD.
This also allows a minor simplification in the Makefile.
|
| | |
| |
| |
| | |
This reverts commit bfc86546849e15dd98362852e76fa9feac2fba77.
|
| |\ \ |
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When running index.sh on OpenBSD, the following error happens for
each item in the news/ directory (output is from "bash -x"):
+ touch -d '4 Jun 2017' news/andrew-robbins-new-maintainer.md
touch: out of range or illegal time specification: YYYY-MM-DDThh:mm:ss[.frac][Z]
This is because OpenBSD's touch(1) requires that the "d" flag's argument
be in ISO 8601 format, that is, "YYYY-MM-DDThh:mm:ss[.frac][Z]". This
could have been dealt with by converting the article date (determined
by "sed -n 3p $f | sed -e 's/^..//g'") to ISO 8601 format, then passing
the date to touch(1). That would have required even more code, so was
discarded as a possible solution.
Instead, this has been solved by prepending the ISO 8601 date to the names
of all news items. This has the benefit of avoiding the need for touch(1)
altogether, as a lexicographic sorting of ISO 8601 dates is the same as
a date-based sorting. In other words, "ls news/*.md" will give a list
of articles sorted by date, which we can then append to news/index.md in
that order.
One downside of this solution is that it introduces the possibility that
the date in the filename (ISO 8601 format) of a news article does not
match the date inside the article (e.g., 1 May 2017). I have not dealt
with this as it remains to be seen whether it will be a problem in
practice.
|
| |\ \ |
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Apparently the name used in a 'for /name/ ...' loop rebinds any local
variables (with the same name) with whatever /name/ is bound
to in a way that persists even after the loop returns. The crux of
the issue here is that a function's children can rebind the parent(s)'
local variables just by using the same name as the variable in a for
loop, which is surprising--and apparently undocumented--behavior.
Use of a subshell group for encapsulating the for loops (See:
project_usage_actions) mitigates the aforementioned issue.
Closes issue: #244
|
| |\ \ \ |
|
| | |/ / |
|
| |\ \ \ |
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This is the same idea as 497a49162f7c8b5c1d7c653087b0ac6c8e5765f9.
Without this patch, the following errors occur when running index.sh:
sed: 1: "news/new-mailing-lists. ...: extra characters at the end of n command
sed: 1: "news/andrew-robbins-new ...: extra characters at the end of n command
sed: 1: "news/formalised-structu ...: extra characters at the end of n command
sed: 1: "news/proposal-rejoin-gn ...: extra characters at the end of n command
sed: 1: "news/unity.bare.html": extra characters at the end of n command
|
| |\ \ \ \
| |_|/ /
|/| | | |
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Before the Makefile, publish.sh was executed on markdown source
files using find(1), which happened like this:
./publish.sh ./index.md
Now that we have a Makefile, this happens instead:
./publish.sh index.md
Note that the file argument "index.md" in the first and second case
both refer to the same file, yet they are different strings. This is
important because publish.sh gives index.md (among other files) special
treatment, and it does this by string comparison.
Unfortunately, only the argument in the first case ("./index.md") will
cause publish.sh to give special treatment, while the argumnent in the
second case ("index.md") will not.
To fix this, make it so that both "./index.md" and "index.md" trigger
publish.sh's special handling. This commit also fixes the same issue
for "docs/fdl-1.3.md" and "conduct.md".
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This is done by replacing www/generate.sh with a Makefile. Benefits:
- Makes builds incremental, meaning that only the minimum number of
markdown files will be converted to HTML during a build. The previous
scheme always generated a new HTML file for every markdown file,
which is a big waste of time if only 1 or 2 markdown files have been
changed.
- Allows for much faster builds through concurrent jobs (e.g., via
"make -j4"). On my 4-core machine, my average build time for the
website with generate.sh was just over 26 seconds; with "make -j4",
it was 13 seconds.
- Avoids portability issues with find(1) in generate.sh, which I was
encountering on OpenBSD.
A note on portability: unlike GNU Make, OpenBSD's Make does not have
the "$(shell [commands])" construct, so we don't use that. Instead we
use "!= [commands]", which is supported by both.
|
| | |/ /
| | |
| | |
| | |
| | |
| | | |
Instead of always using the same file (temp.md), use a unique temporary
file so that multiple instances of publish.sh do not clobber each
other's work.
|
| |\ \ \
| |/ /
|/| | |
|
| |/ / |
|
| |\ \ |
|
| |/ /
| |
| |
| | |
They are generated files, so should probably not be tracked.
|
| |\ \ |
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
On OpenBSD, publish.sh errors out at the following call
to sed(1) (from "bash -x publish.sh"):
+ sed temp.md -i -e 's/\.md\(#[a-z\-]*\)*)/.html\1)/g'
sed: 1: "temp.md": undefined label 'emp.md'
It seems that "temp.md" is being parsed by sed as a sed command,
not as a named file. This is likely due to OpenBSD's strict
usage requirements for sed:
usage: sed [-aEnru] [-i[extension]] command [file ...]
sed [-aEnru] [-e command] [-f command_file] [-i[extension]] [file ...]
As shown above, the sed command must always come before any named
files. This commit does that, which fixes publish.sh with OpenBSD's sed.
This is also tested and working with GNU sed v4.2.2
|
| |\ \ |
|
| |/ /
| |
| |
| |
| |
| |
| | |
OptiPNG losslessly optimizes .png files; in other words, the images
in question should (and do, by my testing) look the same.
In short, we save some bytes for free.
|
| |\ \ |
|
| | | |
| | |
| | |
| | |
| | | |
Be consistent with marking up commands as such, so the documentation
is less surprising.
|
| |\ \ \
| |/ /
|/| | |
|
| | | |
| | |
| | |
| | |
| | | |
The openbsd1 partition is only present for users with an MBR
installation of OpenBSD, which not everyone has.
|
| | | |
| | |
| | |
| | | |
Avoid jargon, add markup where applicable, rewrite for brevity, etc.
|
