Re: index-pack died on pread

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Jul 25, 2007, 23:44 UTC
Message-ID: <alpine.LFD.0.999.0707251636490.3607@woody.linux-foundation.org>
In-Reply-To: <200707260115.13234.robin.rosenberg.lists@dewire.com>

On Thu, 26 Jul 2007, Robin Rosenberg wrote:

> 
> Does cygwin have the same pread problem then.. ? after make NO_PREAD=1 with 1.5.2.4
> clone works.

Interesting.

It's true that pread() is used much less than normal reads, and maybe the cygwin pread() is indeed broken. But it's intriguing how apparently both HP-UX and Cygwin are showing the same breakage.

But if git was doing something odd with pread(), then NO_PREAD shouldn't work either, because that just enables the exact same code, except we now supply a pread emulation library function (which does it using "lseek()": not a *good* emulation, but good enough for the very limited use that git puts it to).

So yeah, it looks like pread() is broken under cygwin and hpux-11.11.

Googling for "pread" "cygwin" does seem to show that it's been buggy at times. Eg:

	Fixes since 1.5.21-1:

...

	- Fix pread in case current file offset is non-zero. (Hideki Iwamoto)

but I'm a bit surprised that hp-ux has the bug. pread isn't *that* complex.

		Linus