From: Nicolas Pitre <nico@fluxnic.net>
To: Junio C Hamano <gitster@pobox.com>
Cc: Lukas Fleischer <lfleischer@lfos.de>,
git@vger.kernel.org,
Johannes Schindelin <Johannes.Schindelin@gmx.de>,
Jeff King <peff@peff.net>
Subject: Re: [PATCH v4] Refactor recv_sideband()
Date: Tue, 28 Jun 2016 17:44:30 -0400 (EDT) [thread overview]
Message-ID: <alpine.LFD.2.20.1606281726330.24439@knanqh.ubzr> (raw)
In-Reply-To: <xmqqwpl96mvv.fsf@gitster.mtv.corp.google.com>
On Tue, 28 Jun 2016, Junio C Hamano wrote:
> Nicolas Pitre <nico@fluxnic.net> writes:
>
> >> There is something else going on. I cannot quite explain why I am
> >> getting this failure from t5401-update-hooks.sh, for example:
> >>
> >> --- expect 2016-06-28 19:46:24.564937075 +0000
> >> +++ actual 2016-06-28 19:46:24.564937075 +0000
> >> @@ -9,3 +9,4 @@
> >> remote: STDERR post-receive
> >> remote: STDOUT post-update
> >> remote: STDERR post-update
> >> +remote: To ./victim.git
> >> not ok 12 - send-pack stderr contains hook messages
> >>
> >> ... goes and looks what v2.9.0 produces, which ends like this:
> >>
> >> ...
> >> remote: STDERR post-receive
> >> remote: STDOUT post-update
> >> remote: STDERR post-update
> >> To ./victim.git
> >> e4822ab..2b65bd1 master -> master
> >> ! [remote rejected] tofail -> tofail (hook declined)
> >>
> >> The test checks if lines prefixed with "remote: " match the expected
> >> output, and the difference is an indication that the new code is
> >> showing an extra incomplete-line "remote: " before other parts of
> >> the code says "To ./victim.git" to report where the push is going.
> >
> > Ah... I think I know what's going on.
> >
> > The leftover data in the strbuf is normally (when there is no errors) an
> > unterminated line. So instead of doing:
> >
> > - fprintf(stderr, "%s: protocol error: no band designator\n", me);
> > + strbuf_addf(&outbuf,
> > + "\n%s: protocol error: no band designator\n",
> > + me);
> >
> > you could omit the final \n in the format string and:
> >
> > - if (outbuf.len > 0)
> > - fprintf(stderr, "%.*s", (int)outbuf.len, outbuf.buf);
> > + if (outbuf.len)
> > + fwrite(outbuf.buf, 1, outbuf.len, stderr);
> > strbuf_release(&outbuf);
> >
> > and here a \n could be added before writing out the buffer.
>
> Unfortunately, that is not it.
>
> The basic structure of the code (without the "SQUASH" we discussed)
> looks like this:
>
> strbuf_addf(&outbuf, "%s", PREFIX);
> while (retval == 0) {
> len = packet_read(in_stream, NULL, NULL, buf, LARGE_PACKET_MAX, 0);
> ...
> band = buf[0] & 0xff;
> switch (band) {
> case 3:
> ... /* emergency exit */
> case 2:
> while ((brk = strpbrk(b, "\n\r"))) {
> int linelen = brk - b;
>
> if (linelen > 0) {
> strbuf_addf(&outbuf, "%.*s%s%c",
> linelen, b, suffix, *brk);
> } else {
> strbuf_addf(&outbuf, "%c", *brk);
> }
> fprintf(stderr, "%.*s", (int)outbuf.len,
> outbuf.buf);
> strbuf_reset(&outbuf);
> strbuf_addf(&outbuf, "%s", PREFIX);
> b = brk + 1;
> }
> if (*b)
> strbuf_addf(&outbuf, "%s", b);
> break;
> ...
> }
> }
>
> if (outbuf.len > 0)
> fprintf(stderr, "%.*s", (int)outbuf.len, outbuf.buf);
>
> Imagine we are reading from band #2 and we find a complete line. We
> concatenate the payload up to the LF at the end of the line to the
> PREFIX we prepared outside the loop and emit it, and then we ASSUME
> that we further have something after strpbrk() and add PREFIX to the
> buffer, before going to the next line in the payload.
>
> But there may not be anything after the LF. outbuf.len is still
> counting the PREFIX and we end up showing it, without termination.
You're right. Although my previous observations still apply.
> This takes us back to what I said in my review of an earlier round,
> in $gmane/297332, where I said:
>
> Instead of doing "we assume outbuf already has PREFIX when we add
> contents from buf[]", the code structure would be better if you:
>
> * make outbuf.buf contain PREFIX at the beginning of this innermost
> loop; lose the reset/addf from here.
>
> * move strbuf_reset(&outbuf) at the end of the next if (*b) block
> to just before "continue;"
>
> perhaps?
>
> I think the strbuf_addf(PREFIX) above the loop should be removed,
> and instead the code should use the PREFIX only when it decides that
> there is something worth emitting, i.e.
>
> while (!retval) {
> len = packet_read();
> ...
> band = buf[0] & 0xff;
> switch (band) {
> case 3:
> ... /* emergency exit */
> case 2:
> while ((brk = ...)) {
> /* we have something to say */
> strbuf_reset(&outbuf);
> strbuf_addstr(&outbuf, PREFIX);
That won't work. If at this point there is the beginning of a partial
line queued in the buffer from the previous round waiting to see its
line break, you just deleted the beginning of that line.
Furthermore, that partial line won't get a prefix if it doesn't have at
least one line break in the packet data.
Rather the prefix should be added whenever the buffer is empty before
every addition.
> if (linelen)
> strbuf_addf(...);
> else
> strbuf_addch(*brk);
> fwrite(outbuf.buf, 1, outbuf.len, stderr);
> b = brk + 1;
> }
> if (*b) {
> /* we still have something to say */
> strbuf_reset(&outbuf);
> strbuf_addstr(&outbuf, PREFIX);
> strbuf_addf(...);
This is also wrong. If the middle part of a partial line is received,
you just deleted its queued beginning.
Nicolas
next prev parent reply other threads:[~2016-06-28 21:44 UTC|newest]
Thread overview: 60+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-06-13 19:52 [PATCH] Refactor recv_sideband() Lukas Fleischer
2016-06-13 21:07 ` Nicolas Pitre
2016-06-14 13:44 ` Johannes Schindelin
2016-06-14 15:04 ` Nicolas Pitre
2016-06-14 15:30 ` Johannes Schindelin
[not found] ` <Cq7rbYgOpb0CVCq7sbGmpL@videotron.ca>
2016-06-14 16:43 ` Nicolas Pitre
2016-06-14 17:09 ` Nicolas Pitre
[not found] ` <CsLdb3qLMBok7CsLebwX38@videotron.ca>
2016-06-14 17:55 ` Nicolas Pitre
2016-06-14 18:11 ` Junio C Hamano
2016-06-14 19:11 ` Lukas Fleischer
2016-06-14 19:16 ` Junio C Hamano
[not found] ` <Ct7VbfLfTHEALCt7Wbh8Xs@videotron.ca>
2016-06-14 20:10 ` Nicolas Pitre
2016-06-14 21:00 ` [PATCH v2] " Lukas Fleischer
2016-06-14 21:11 ` Lukas Fleischer
2016-06-14 21:25 ` Junio C Hamano
2016-06-15 3:44 ` Jeff King
[not found] ` <146597489449.32143.1327156804178869158@s-8d3a2dc3.on.site.uni-stuttgart.de>
2016-06-19 10:48 ` Lukas Fleischer
2016-06-24 15:31 ` Jeff King
2016-06-24 17:45 ` Johannes Schindelin
2016-06-24 18:14 ` Jeff King
2016-06-24 18:32 ` Junio C Hamano
2016-06-27 10:58 ` Lukas Fleischer
2016-06-27 15:54 ` Junio C Hamano
2016-06-27 16:16 ` Jeff King
2016-06-27 17:50 ` Junio C Hamano
2016-06-27 20:34 ` Lukas Fleischer
2016-06-27 20:47 ` Nicolas Pitre
2016-06-28 4:01 ` Lukas Fleischer
2016-06-28 5:20 ` Junio C Hamano
2016-06-28 10:04 ` Johannes Schindelin
2016-06-28 10:05 ` Johannes Schindelin
2016-06-28 15:13 ` Junio C Hamano
2016-06-28 16:21 ` Johannes Schindelin
2016-06-24 20:07 ` Dennis Kaarsemaker
2016-06-22 5:29 ` [PATCH v3] " Lukas Fleischer
2016-06-22 15:02 ` Nicolas Pitre
2016-06-22 22:47 ` Nicolas Pitre
2016-06-23 17:35 ` Lukas Fleischer
2016-06-23 18:59 ` Nicolas Pitre
2016-06-28 4:35 ` [PATCH v4] " Lukas Fleischer
2016-06-28 16:57 ` Junio C Hamano
2016-06-28 17:24 ` Junio C Hamano
2016-06-28 17:46 ` Nicolas Pitre
2016-06-28 18:13 ` Junio C Hamano
2016-06-28 18:28 ` Nicolas Pitre
2016-06-28 19:51 ` Junio C Hamano
2016-06-28 20:36 ` Nicolas Pitre
2016-06-28 21:09 ` Junio C Hamano
2016-06-28 21:44 ` Nicolas Pitre [this message]
2016-06-28 22:33 ` Junio C Hamano
2016-06-28 22:47 ` Junio C Hamano
2016-06-29 3:00 ` Junio C Hamano
2016-06-29 3:41 ` Nicolas Pitre
2016-06-29 2:02 ` Nicolas Pitre
2016-06-29 16:40 ` Junio C Hamano
2016-06-30 6:16 ` Lukas Fleischer
2016-07-01 20:01 ` Junio C Hamano
2016-07-05 20:35 ` Nicolas Pitre
2016-07-06 21:11 ` Junio C Hamano
2016-07-07 0:56 ` Nicolas Pitre
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.LFD.2.20.1606281726330.24439@knanqh.ubzr \
--to=nico@fluxnic.net \
--cc=Johannes.Schindelin@gmx.de \
--cc=git@vger.kernel.org \
--cc=gitster@pobox.com \
--cc=lfleischer@lfos.de \
--cc=peff@peff.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).