Re: [PATCH v2] Refactor recv_sideband()
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Jun 27, 2016, 15:54 UTC
- Message-ID
- <xmqqr3bibpap.fsf@gitster.mtv.corp.google.com>
- In-Reply-To
- <146702508453.24123.590646528169139972@s-8d3a37fa.on.site.uni-stuttgart.de>
Lukas Fleischer <lfleischer@lfos.de> writes:
Show 15 quoted lines
> On Fri, 24 Jun 2016 at 20:32:28, Junio C Hamano wrote: >> On Fri, Jun 24, 2016 at 11:14 AM, Jeff King <peff@peff.net> wrote: >> > >> > I do wonder about the ANSI_SUFFIX thing, though. >> >> compat/winansi.c seems to have a provision for 'K' (and obviously 'm' >> for colors). > > So we probably want to switch back to using fprintf() or fputs() and > rely on the assumption that these functions use atomic write() > operations internally when printing to stderr (on practically relevant > systems, i.e., libc implementations), right? > > Another option is using write() on all systems but Windows and fall back > on fputs() but that feels a bit like over-engineering at this point.
Wait a bit.
The original used to use one fprintf(stderr) per line it outputs. That does not necessarily guarantee atomicity (i.e. the payload string may be large and make stdio decide to do multiple writes).
The issue found in your first round (sent on Jun 13) was that you did multiple fprintf(stderr) when showing a single line, which is guaranteed to result in multiple writes because stderr is not line-buffered. But that was corrected in your second round.
It's just you used xwrite() there that introduced a different issue. Wouldn't replacing it with fwrite(stderr) without changing anything else solve that?