190863 - some Images fail to load with pipelining enabled

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect, P1)

Description

[Matthias Versen [:Matti]]

win2k and build 20030127.. (cvs trunk) and Kairo saw this also with linux and
20030126..

1) load the URL
2) use shift+reload (the images are possible already in the cache due to chatzilla)
3) random images aren't loading or only a part is loaded
4) shift+reload and different images aren't loading

works with disabled pipelining and also with 1.3a and pipelining enabled

Comment 1

[Matthias Versen [:Matti]]

Attachment: Zipped httplog

Comment 2

[Matthias Versen [:Matti]]

Attachment: screenshot from the page while generating the logfile

Comment 3

[Bernard Alleysson]

Some images fail to load with pipelining enabled on http://www.tf1.fr/news
I don't know if this is the same bug though

Comment 4

[Darin Fisher]

Attachment: v1 patch

turns out the async-io patch caused this regression.  when writing out the
request, we would fail if the transaction did not write all that it said it
could write provided the socket did not fail with a would block error.	this
causes problems with the way pipelined requests are written out in some cases. 
the fix is to clean up the way we write to the sockets.  the code to write to
the sockets now parallels the code to read from the sockets, which is nice.

Comment 5

[Darin Fisher]

this patch depends on the patch for bug 190549.

Comment 6

[Darin Fisher]

Attachment: v1.1 patch (the last patch had some junk in it)

the last patch had some junk in it...

Comment 7

[Darin Fisher]

i think this should block 1.3 beta because it breaks pipelining.

Comment 8

[Darin Fisher]

another testcase is http://www.spartanbookstore.com/

Comment 9

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 112886 [details] [diff] [review]
v1.1 patch (the last patch had some junk in it)

>+            // if the transaction didn't want to write anymore data, then

"any more"

>+        // write more from the socket until error or end-of-request...

"to the socket"

Comment 10

[Darin Fisher]

typos fixed in my local tree.

Comment 11

[Darin Fisher]

Comment on attachment 112886 [details] [diff] [review]
v1.1 patch (the last patch had some junk in it)

request drivers approval to land fix for this bug.  relatively low-risk.

Comment 12

[Asa Dotzler [:asa]]

Comment on attachment 112886 [details] [diff] [review]
v1.1 patch (the last patch had some junk in it)

a=asa (on behalf of drivers) for checkin to 1.3beta.

Comment 13

[Darin Fisher]

fixed-on-trunk

Comment 14

[Olivier Cahagne]

*** Bug 191082 has been marked as a duplicate of this bug. ***