sendfile test errors

[Mario Domenech Goulart]

sendfile's test suite shows errors on different platforms:

• linux/x86 (chicken 4.7.0.3-st): ​http://paste.call-cc.org/paste?id=75b3ffa3059d572fadd4fb49b094868dc9828871

• linux/x86-64 (chicken 4.7.0): ​http://paste.call-cc.org/paste?id=75b3ffa3059d572fadd4fb49b094868dc9828871#a1

• macoxs/x86-64 (master branch): ​http://tests.call-cc.org/master/macosx/x86_64/2011/11/14/salmonella-report/test/sendfile.html

• linux/x86 (master branch): ​http://tests.call-cc.org/master/linux/x86/2011/11/17/salmonella-report/tests/sendfile.html

[Jim Ursetto]

Also, sending a file on macosx 64-bit 10.7 with sendfile results in corruption at byte 81661 (I have also seen 83K).

[Jim Ursetto]

I believe the problem on OS X is that sendfile() will return EAGAIN even if it was able to send partial data; this is then discarded by the implementation, which assumes zero bytes were sent and tries to retransmit the full buffer.

[Jim Ursetto]

This patch fixes the corruption on OS X. Use hg import to import it, I do not feel like forking on bitbucket so sue me lolololol

[Jim Ursetto]

Fix corruption on OS X related to EAGAIN

[certainty]

Thanks for reporting this, Jim. There are some tests which fail occasionally as well.
I don't know why yet. I'll happily apply your patch

[certainty]

I've applied the patch to trunk and will soon tag a release

[felix winkelmann]

Milestone 4.8.0 deleted