Eric Wong [Thu, 19 Oct 2023 01:14:31 +0000 (01:14 +0000)]
lei: simplify startq/au_done wakeup notifications
We only need to write one byte at MUA start instead of a byte
for every LeiXSearch worker. Also, make sure it succeeds by
enabling autodie for syswrite.
When reading, we can rely on `:perlio' layer `read' semantics
to retry on EINTR to avoid looping and other error checking.
Eric Wong [Tue, 17 Oct 2023 23:38:05 +0000 (23:38 +0000)]
test_common: only hide TCP port in messages
v2:// lei outputs are on the filesystem, so putting $HOST:$PORT
is nonsensical. We'll also keep `127.0.0.1' or `[::1]' since
it's harmless and can point out obvious errors in system
configuration when testing with old Perls or libraries.
Eric Wong [Tue, 17 Oct 2023 23:37:58 +0000 (23:37 +0000)]
xt/git-http-backend: remove Net::HTTP usage
HTTP::Tiny is part of the Perl standard library since Perl 5.14
while Net::HTTP has never been (unlike Net::NNTP or Net::POP3).
For the test which forces server-side buffering, we'll just use
regular socket handle.
Eric Wong [Tue, 17 Oct 2023 23:37:54 +0000 (23:37 +0000)]
xap_helper: die more easily in both implementations
We don't need to tolerate bad requests since it's only handling
requests from the parent process. So simplify error management
and just die||exit if we get a bad request.
Eric Wong [Tue, 17 Oct 2023 23:37:52 +0000 (23:37 +0000)]
use read_all in more places to improve safety
`readline' ops may not detect errors on partial reads.
This saves us some code to reduce cognitive overhead for
readers. We'll also support reusing a destination buffers so it
can work more nicely with existing code.
Eric Wong [Tue, 17 Oct 2023 23:37:49 +0000 (23:37 +0000)]
git: introduce read_all function
This makes it easier to improve error checking, since the
`do { local $/; readline(FH) }' construct does not detect
errors (autodie does not cover `readline' or `<FH>').
I'm not sure exactly where this should be, but PublicInbox::Git
is used nearly everywhere in our code base and it's probably
not worth creating a new package for it.
Eric Wong [Tue, 17 Oct 2023 23:37:46 +0000 (23:37 +0000)]
lei_mirror: start converting to autodie
This code is too noisy and not critical for startup performance;
so autodie provides a nice noise reduction while improving error
reporting in most cases.
For places where failures are expected, the `CORE::' prefix
gives us an easy escape hatch to fall back to normal error
checking.
Eric Wong [Tue, 17 Oct 2023 10:11:06 +0000 (10:11 +0000)]
input_pipe: handle noncanonical TTY
lei could get a TTY in noncanonical mode for stdin, so rely on
VMIN+VTIME to get the desired non-blocking semantics we'd expect
from a pipe or socket. This ought to prevent read(2) (Perl sysread)
from returning zero when we really want to hit EAGAIN.
Eric Wong [Tue, 17 Oct 2023 10:11:05 +0000 (10:11 +0000)]
input_pipe: improve error handling
Ensure the callback is always guarded by `eval' to catch
exceptions and to force a ->close (EPOLL_CTL_DEL).
We also don't want to blindly set O_NONBLOCK on TTYs since their
O_NONBLOCK semantics aren't well-defined by POSIX. We can also
drop EPOLLET (edge-triggered) use to reduce the need to make
->requeue calls on our end.
Eric Wong [Tue, 17 Oct 2023 10:11:04 +0000 (10:11 +0000)]
lei: consolidate stdin slurp, fix warnings
We can share more code amongst stdin slurper (not streaming)
commands. This also fixes uninitialized variable warnings when
feeding an empty stdin to these commands.
Eric Wong [Sun, 15 Oct 2023 08:16:28 +0000 (08:16 +0000)]
learn: respect indexlevel for v1 inboxes
v2 never suffered from this bug, apparently, but -learn didn't
seem able to handle indexlevel=basic (nor respect `medium')
for v1 inboxes. I only noticed this bug because I converted
some ancient v1 inboxes to `basic' to save space.
Eric Wong [Fri, 13 Oct 2023 06:12:29 +0000 (06:12 +0000)]
xap_helper_cxx: allow sharing XDG_CACHE_HOME across ABIs
For users sharing home directories (or just XDG_CACHE_HOME)
across hosts of different architectures, we must use a compiler
and architecture-specific destination directory for storing the
binary result. Even on the same OS and architecture, different
C++ compilers may have different ABIs, so we must account for
that.
Eric Wong [Thu, 12 Oct 2023 00:21:00 +0000 (00:21 +0000)]
lei: quiet excessive write/seen messages
We don't want to end up dumping nr_seen/nr_write when progress
is disabled, nor do we want forked off `lei note-event' workers
dump them when DS->Reset is called on fork.
Eric Wong [Wed, 11 Oct 2023 07:20:57 +0000 (07:20 +0000)]
lei import|tag|rm: support --commit-delay=SECONDS
Delayed commits allows users to trade off immediate safety for
throughput and reduced storage wear when running multiple
discreet commands.
This feature is currently useful for providing a way to make
t/lei-store-fail.t reliable and for ensuring `lei blob' can
retrieve messages which have not yet been committed.
In the future, it'll also be useful for the FUSE layer to batch
git activity.
Eric Wong [Wed, 11 Oct 2023 07:20:55 +0000 (07:20 +0000)]
import: cat_blob is a no-op w/o live fast-import
cat_blob is a fallback for handling files which haven't made it
onto disk to be readable by `git cat-file'. Thus spawning a new
fast-import process to retrieve a blob is pointless, as cat_blob
is only used as a last resort when `git cat-file' fails.
Eric Wong [Wed, 11 Oct 2023 07:20:54 +0000 (07:20 +0000)]
import: switch to Unix stream socket for fast-import
We use fewer file descriptors and fewer lines of code this way.
I'm not aware of any place we rely on POSIX pipe semantics with
`git fast-import', and sockets have bigger buffers by default
in most cases (even if Linux allows larger pipe buffers).
Eric Wong [Wed, 11 Oct 2023 07:20:53 +0000 (07:20 +0000)]
treewide: consolidate "From " line removal
Aside from our prior import bugs (fixed in a0c07cba0e5d8b6a
(mda: drop leading "From " lines again, 2016-06-26)), we'll
always have to be dealing with mutt piping messages to us and
`git format-patch' output. So just share the regexp so we
can use it everywhere.
In may be desirable to allow importing messages with a leading
"From " line for FUSE, even.
Additionally, some instances of this regexp needlessly added
optional `\r?' (CR) checks ahead of the `\n' (LF) element; but
they're pointless anyways since [^\n]* is enough to exclude all
non-LF bytes.
Eric Wong [Wed, 11 Oct 2023 07:20:51 +0000 (07:20 +0000)]
msgtime: quiet warnings we can do nothing about
In retrospect, warning about bad times and dates is pointless
since there's nothing actionable about it. We'll also drop an
unnecessary capture in msg_received_at while we're at it and
favor using $eml since as the input variable name to match
current usage.
The note to install Date::Parse as a fallback remains since it
can be helpful in some cases (and is actionable by the user).
Eric Wong [Wed, 11 Oct 2023 07:20:50 +0000 (07:20 +0000)]
lei_xsearch: improve curl progress reporting
Instead of having tail(1) follow a file when we're in verbose
mode, unconditionally pipe stderr to a Perl 2-liner which tees
its output to a regular file with line buffering.
POSIX tee(1) isn't suitable for this task since it's required
to be completely unbuffered while we want line-buffering when
running parallel processes. Fortunately, Perl makes this easy.
This also means we no longer leave curl-err.XXXX files around
on premature shutdown if we're hit by a SIGKILL or similar and
can't exit normally.
We do need to stop and respawn the Perl process if we hit a curl
error, though, since we need to be certain the output is
flushed.
Eric Wong [Wed, 11 Oct 2023 07:20:49 +0000 (07:20 +0000)]
lei rediff: use ProcessIO for --drq support
This required fixing binmode support a few commits ago, along
with properly enabling autoflush in popen_wr instead of setting
it on the wrapper ProcessIO class.
Eric Wong [Tue, 10 Oct 2023 10:09:04 +0000 (10:09 +0000)]
ds: use a dummy poller during Reset
commit 1897c3be1ed644a05f96ed06cde4a9cc2ad0e5a4
(ds: Reset: replace Poller object early, 2023-10-04)
was not effective at eliminating the following message
at daemon shutdown:
Can't call method "FILENO" on an undefined value at
.../PublicInbox/Select.pm line 34 during global destruction.
This seems down to some tied objects having unpredictable
destruction order. So use a dummy class to ensure its ep_*
methods never call the tied `FILENO' method at all since
dropping the Poller object will release any resources it holds.
Eric Wong [Tue, 10 Oct 2023 10:07:56 +0000 (10:07 +0000)]
over*: avoid defined-or hash assignments with side-effects
These may've been causing strange errors[1] in t/imapd.t from
the -watch daemon, such as:
Cannot copy to HASH in scalar assignment ../PublicInbox/Over.pm
in the Over->dbh() sub. I've only noticed this failure on
FreeBSD 13.2 (Perl 5.32.1, DBD::SQLite 1.72 (bundled SQLite
3.39.4), DBI 1.643) so far, so it could also be something to do
with the versions used and/or memory layout differences
with libc or build toolchain.
Eric Wong [Tue, 10 Oct 2023 09:03:09 +0000 (09:03 +0000)]
t/nntp.t: attempt to track source of undefined vars
Occasionally, t/nntp.t spews undefined variable warnings under
`make check-run'. While the test doesn't fail, it's annoying
to see them and it could be a source of deeper problems.
Eric Wong [Mon, 9 Oct 2023 17:56:23 +0000 (17:56 +0000)]
www_coderepo: fix handling of non-UTF-8 git data
We can't assume git output is UTF-8, and we'll always have
legacy data in git coderepos. So attempt to display some
some garbled text rather than nothing at all if Perl croaks
on it.
sox commit c38987e8d20505621b8d872863afa7d233ed1096
(Added raw inverse-bit u-law and A-law support. Updated *.txt files., 2001-12-13)
is an example of a commit which caused problems for me.
Eric Wong [Sun, 8 Oct 2023 22:11:48 +0000 (22:11 +0000)]
introduce ProcessIONBF for multiplexed non-blocking IO
This is required for reliable epoll/kevent/poll/select
wakeup notifications, since we have no visibility into
the buffer states used internally by Perl.
We can safely use sysread here since we never use the :utf8
nor any :encoding Perl IO layers for readable pipes.
I suspect this fixes occasional failures from t/solver_git.t
when retrieving the WwwCoderepo summary.
Eric Wong [Sun, 8 Oct 2023 22:11:47 +0000 (22:11 +0000)]
process_io: fix binmode and use it in lei_xsearch
The `binmode' perlop can only take two scalars, so passing
`@_' blindly won't work since prototypes are checked. This
means we can get IO::Uncompress::Gunzip working properly
with ProcessIO and use it for curl.
We'll also just autodie (instead of warn) on FS errors when
dealing with curl stderr; since the process will likely be
in bigger trouble soon, anyways.
Eric Wong [Sun, 8 Oct 2023 20:19:40 +0000 (20:19 +0000)]
overidx: use croak/confess instead of die
Unlike `die', `croak' can be expanded to `confess' to give a
full backtrace. We'll use `confess' on transaction failures
since that occasionally causes sporadic t/imapd.t failures on
FreeBSD (IO::Kqueue is installed, so signals are deferred).
Eric Wong [Sat, 7 Oct 2023 21:24:09 +0000 (21:24 +0000)]
process_io: pass args to awaitpid as list
Specifying {cb_args} in the options hash felt awkward to me.
Instead, just use the Perl stack like we do with awaitpid()
and pass the list down directly.
Eric Wong [Sat, 7 Oct 2023 21:24:08 +0000 (21:24 +0000)]
rename ProcessPipe to ProcessIO
Since we deal with pipes (of either direction) and bidirectional
stream sockets for this class, it's better to remove the `Pipe'
from the name and replace it with `IO' to communicate that it
works for any form of IO::Handle-like object tied to a process.
Eric Wong [Sat, 7 Oct 2023 21:24:07 +0000 (21:24 +0000)]
import: use autodie, rely on PerlIO for retries
As documented in perlipc(1), the default :perlio layer retries
the `read' perlop on EINTR. The :perlio layer also makes `read'
perform read-in-full behavior; so there's no need to loop
ourselves. Our responsibility is now only to detect short reads
in case fast-import is killed mid-stream.
Eric Wong [Sat, 7 Oct 2023 21:24:06 +0000 (21:24 +0000)]
ipc: use autodie for most syscalls
I'm not sure how/if we should bother recovering from these,
so just croak and let some caller deal with it. `autodie'
uses Carp internally, so setting `PERL5OPT=-MCarp=verbose'
in the environment gives us full stacktraces.
Eric Wong [Sat, 7 Oct 2023 21:24:05 +0000 (21:24 +0000)]
ipc: require fork+SOCK_SEQPACKET for wq_* functions
None of the lei internals works properly without forking and
sockets. The fallback code increases the potential to accidentally
call subs in the wrong process during the teardown phase.
We'll still support ipc_do w/o forking for now since it
forking doesn't benefit small indexing runs from -mda and
such.
Eric Wong [Sun, 8 Oct 2023 05:49:34 +0000 (05:49 +0000)]
lei: fix implicit stdin support for pipes
Eric Wong <e@80x24.org> wrote:
> +++ b/t/lei-store-fail.t
> + my $cmd = [ qw(lei import -q -F mboxrd) ];
> + my $tp = start_script($cmd, undef, $opt);
Of course the lack of `-' or `--stdin' only worked on Linux and
NetBSD, but not other BSDs.
-------8<------
Subject: [PATCH] lei: fix implicit stdin support for pipes
st_mode permission bits can't be used to determine if a file or
pipe we have on stdin readable or not. Writable regular files
can be opened O_RDONLY, and permissions bits for pipes are
inconsistent across platforms.
On FreeBSD, OpenBSD, and Dragonfly, only the S_IFIFO bit is set
in st_mode with none of the permission bits are set. Linux and
NetBSD have both the read and write permission bits set for both
ends of a the pipe, so they're just as inaccurate but allowed
the feature to work before this change.
For now, we'll just assume our users know that stdin is intended
for input and consider any pipe or regular file to be readable.
If we were to be pedantic, we'd check O_RDONLY or O_RDWR
description flags via the F_GETFL fcntl(2) op to determine if a
pipe or socket is readable. However, I don't think it's worth
the code to do so.
Eric Wong [Fri, 6 Oct 2023 09:46:01 +0000 (09:46 +0000)]
xap_helper.h: strdup keys for DragonFlyBSD hdestroy(3)
DragonFlyBSD matches OpenBSD behavior in freeing every single key on
hdestroy(3). I suppose hdestroy(3) is neglected enough these days that
nobody cares and we'll likely introduce a small C hash table such as
khash (also used within git).
Eric Wong [Fri, 6 Oct 2023 09:46:00 +0000 (09:46 +0000)]
kqnotify: drop EV_CLEAR (edge triggering)
I'm not entirely certain how it works with the way we use
kevent. I do know IO::KQueue has hard-coded kevent retrievals
to 1000 events so it's conceivable we'd end up missing wakeups
as we don't loop or requeue in callers. So just rely on the
*BSD kernel to provided requeue behavior for us by using
level-triggering.
In any case, this seems to workaround t/dir_idle.t failures
on Dragonfly due to a tmpfs bug in all versions up to v6.4.
This ensures script/lei $send_cmd usage is EINTR-safe (since
I prefer to avoid loading PublicInbox::IPC for startup time).
Overall, it saves us some code, too.
Eric Wong [Wed, 4 Oct 2023 17:46:35 +0000 (17:46 +0000)]
makefile: symlink-install: do not depend on realpath
For the Makefile, we can use $(PWD) make macro from make(1posix)
as POSIX requires all environment variables be accessible as
macros, and $PWD is a standard sh(1) environment variable.
lei.sh must quiet the stderr of realpath before falling back to
readlink(1) which is available on NetBSD.
Eric Wong [Wed, 4 Oct 2023 08:50:58 +0000 (08:50 +0000)]
ds: make %AWAIT_PIDS a hash, not hashref
This is more persistent than some of the others and we don't
swap it on use (unlike $nextq or $ToClose). In other words,
it's helpful for communicating its lifetime expectancy is
close to %DescriptorMap and not like to queue-type things
such as $ToClose.
Eric Wong [Wed, 4 Oct 2023 08:50:57 +0000 (08:50 +0000)]
ds: cleanup fork + Reset support
We used to have many entries for %Stack, but nowadays it's just
the one used by next_tick, so just replace it a $cur_runq variable.
I'm reducing reliance on hash keys for things with global scope
to ensure typos can be detected (strict||v5.12 forces us to fix
uses of undeclared variables, but they can't detect typos in
hash keys.
Eric Wong [Wed, 4 Oct 2023 08:50:56 +0000 (08:50 +0000)]
ds: Reset: replace Poller object early
Process shutdown can be chaotic and unpredictable. Try to make
it more predictable by ensuring any PublicInbox::Select object
can't hold references to any objects.
This should fix the following error I saw in syslog during a deploy:
Can't call method "FILENO" on an undefined value at
.../PublicInbox/Select.pm line 34 during global destruction.
Replacing $Poller with PublicInbox::Select (instead of undef-ing
it) means we can avoid adding branches to ->epwait and ->close
before calls to ->ep_mod and ->ep_del, respectively.
Eric Wong [Wed, 4 Oct 2023 03:49:20 +0000 (03:49 +0000)]
lei: get rid of l2m_progress PktOp callback
We already have an ->incr callback we can enhance to support
multiple counters with a single request. Furthermore, we can
just flatten the object graph by storing counters directly in
the $lei object itself to reduce hash lookups.