Date | Commit message (Collapse) |
|
Hackers modifying test cases or manpages still need to use
GNU make, but this should make life easier for folks that
just want to run cmogstored and maybe make minor changes
to the code.
|
|
We depend on GNU make anyways, so use a maintainer-only
Makefile snippet to do proper dependency ordering for
generating the manpage using help2man. The -include
directive requires GNU make, and GNU make runs on more
platforms than we'll ever care to support.
|
|
Ugh, recursive make sucks...
|
|
This seems to make code generation much faster for
http_parser.rl (without modifying its output).
|
|
BUILT_SOURCES isn't required and gives us build-dependency
issues because updating C sources will update .Po files.
|
|
We don't need sleep() at all, and we rely on GCC 4.1+ anyways
so we know alloca() is supported...
|
|
They always go to a log file (via automake) anyways, so there's
no point in hiding information if we ever want to look into a
test failure.
|
|
|
|
In case we publish mirrors to multiple sites.
|
|
We don't want "in /path/to/foo" messages in our output
(NEWS/ChangeLog/NEWS.atom.xml)
|
|
Ugh, looks like I need to fix some mogadm bugs
|
|
|
|
Previous versions had a buggy WEBrick that didn't set
Content-Range in responses correctly.
|
|
Yes we can use kqueue!
|
|
Some filesystems (e.g. ext4) support fractional seconds
in mtime, but HTTP date headers cannot conveying that
portably.
|
|
We support it since all the kqueue issues got worked out.
|
|
Harmless, but spotted by clang.
|
|
This stuff breaks more than code!
|
|
A followup to commit a2c95dd859813d902a7642ad0ca3b41baa4ea748
|
|
I forget my VMs are slower and run too many tests in parallel :x
|
|
iostat doesn't actually show I/O statistics on Debian GNU/kFreeBSD
|
|
The "usage" file may get overwritten by the
usage file generator.
|
|
BUFSIZ is only 1024 on FreeBSD, this is too small to be
optimal for large I/O operations.
|
|
We may just request a bigger stack in the future and
switch to alloca() entirely.
|
|
We need to use NOTE_TRIGGER to wakeup when pushing a mog_fd
into to the active queue of another thread. This is probably
because we need to rely on an "on ready" event for edge-triggered
notifications.
|
|
Oops, there is a trailing space in the output.
|
|
We shouldn't push into the idle queue until we get EAGAIN.
|
|
The tarball name can change during the release check,
so make sure we get the up-to-date name.
|
|
This means we won't have to defer buffers, advance lines
nor need do any buffering. This greatly reduces the
chance of hitting bugs from shuffling/moving buffers
around.
|
|
This should really be tunable, but we can do that later.
|
|
Needless indirection since type errors are pretty easy to
avoid with our code base.
|
|
http->range_bytes_tip is an abandoned idea for reusing
the HTTP request header in the response. Drop that and
rely on Ragel to bail on bad Range requests.
|
|
bin_PROGRAMS (cmogstored) must be up-to-date when firing Perl
or Ruby tests.
|
|
We already default to using -ggdb3 for all builds
|
|
No point in making our own names if the Makefile
provides it for us.
|
|
Hopefully this will encourage people to try this.
|
|
We need a reliable way to ensure Content-Range uploads
are working properly in a client implemented by someone
else.
|
|
By removing objects instead of touching source to force
rebuilds, we can avoid unnecessary warnings if a hacker
has a file open in their $EDITOR.
|
|
This should be reasonably portable, but I'd rather just
support GNU make...
|
|
Should keep the top-level directory cleaner.
|
|
We want as much debugging information as possible, and -ggdb3
should give us even macro expansion.
|
|
It'll be easier to mentally parse out which flags are
_different_.
|
|
|
|
It was a relic from when we couldn't utilize the gmake
jobserver for individual Ruby tests.
|
|
Oops, found by "make release-check"
|
|
Fortunately we only deal with LANs.
|
|
They're far too common and will just flood syslog
|
|
This is to maintain consistency with the Perl mogstored
implementation (and probably most other HTTP daemons out
there).
|
|
This allows clients to support partial requests and
resume downloads.
|
|
It doesn't seem to work reliably in cross-thread wakeup
situations.
|