timeless Reporter
	Description • 22 years ago

SunOS DEATHSTARII 5.8 Generic_108528-17 sun4u sparc SUNW,Sun-Blade-100
Apps listed below live in ~/bin and have their libraries in ~/lib
gmake: GNU Make 3.80
cvs: Concurrent Versions System (CVS) 1.11 (client/server)
glib: 1.2.10
libIDL: 0.6.8
gtar: tar (GNU tar) 1.13.19

Not Installed:
gtk, qt

I'm starting to play with mozilla on solaris (again).
The current configuration is xpcom standalone (I'll attach my mozconfig).

Anyway
$ ./run-mozilla.sh -g -d dbx ./TestPipes
Type Manifest File:
/export/home/timeless/mozilla/debug-xpcom-sparc-sun-solaris2.8/dist/bin/components/xpti.dat
TestChainedPipes
wrote total of 595886 bytes
EOF count = 595886
read  595886 bytes, time = 4484ms
Testing New Pipes: segment size 16 buffer size 16
Testing long writes...
read  1222211 bytes, time = 12226ms
wrote 1222211 bytes, time = 12226ms
Testing short writes...
###!!! ASSERTION: received wrong amount: 'received == writeCount', file
/export/home/timeless/mozilla/xpcom/test
s/TestPipes.cpp, line 270
Break: at file /export/home/timeless/mozilla/xpcom/tests/TestPipes.cpp, line 270
Suspending process; attach with the debugger.
Current function is TestShortWrites
  270           NS_ASSERTION(received == writeCount, "received wrong amount");
(dbx) where
current thread: t@1
  [1] _kill(0x0, 0x17, 0xff35e000, 0x17, 0x656e6400, 0x53750000), at 0xff29f7cc
  [2] nsDebug::Break(0x17457, 0x10e, 0xff2bfc90, 0xff2bfc90, 0x17440, 0x17457),
at 0xff1219f0
  [3] nsDebug::Assertion(0x1742a, 0x17440, 0x17457, 0x10e, 0x3039, 0xffff8000),
at 0xff121838
=>[4] TestShortWrites(in = 0x55bfc, out = 0x55c20), line 270 in "TestPipes.cpp"
  [5] RunTests(segSize = 16U, segCount = 1U), line 541 in "TestPipes.cpp"
  [6] main(argc = 1, argv = 0xffbefaac), line 645 in "TestPipes.cpp"
(dbx) print received
received = 16U
(dbx) print writeCount
writeCount = 17U

The assert happens a *lot*.

	Darin Fisher
	Comment 2 • 22 years ago

probably fallout from my checkin for bug 176919.

-> me

	timeless Reporter
	Comment 3 • 22 years ago

fwiw, while the specific instance here shows received/writecount off by one this
isn't always the case.

	Alex Vincent [:WeirdAl]
	Comment 4 • 16 years ago

I just hit this on Firefox 3 RC2, WinXP, running make check:

TestPipes!RunTests+0x0000000000000160 (c:\gecko1_9\mozilla\xpcom\tests\testpipes.cpp, line 422)
TestPipes!main+0x000000000000009B (c:\gecko1_9\mozilla\xpcom\tests\testpipes.cpp, line 524)
TestPipes!__tmainCRTStartup+0x00000000000001A6 (f:\sp\vctools\crt_bld\self_x86\crt\src\crtexe.c, line 597)
TestPipes!mainCRTStartup+0x000000000000000D (f:\sp\vctools\crt_bld\self_x86\crt\src\crtexe.c, line 414)
kernel32!RegisterWaitForInputIdle+0x0000000000000049

	Alex Vincent [:WeirdAl]
	Comment 5 • 16 years ago

I've hit it again, Vista, FF3, running make check.

	Alex Vincent [:WeirdAl]
	Comment 6 • 16 years ago

Clarification - I hit this every single time I run make check.  received is 16, writeCount is higher, usually equal to 19.

Is this test still valid?

	Alex Vincent [:WeirdAl]
	Comment 7 • 16 years ago

After further examination, I think I'm simply hitting a racing condition.  The segment size is 16, and I notice in my debug output that just as the assertion failure hits, we're reading another few bytes at the same time.

Also, by changing one line:
http://hg.mozilla.org/mozilla-central/index.cgi/annotate/tip/xpcom/tests/TestPipes.cpp#l266
-        if (gTrace)
+        if (PR_TRUE)

The test passes.

	Dimitrios Apostolou
	Comment 8 • 16 years ago

Happens consistently on Irix too, and the above one line patch doesn't seem to remedy the situation. In my case it's received==16, writeCount==17, on the third iteration of the loop.

	Serge Gautherie (:sgautherie) Assignee
	Comment 9 • 16 years ago

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1a2pre) Gecko/2008073018 SeaMonkey/2.0a1pre] (home made) (W2Ksp4)

{{
Testing short writes...
###!!! ASSERTION: received wrong amount: 'received == writeCount', file .../mozilla/xpcom/tests/TestPipes.cpp, line 273

 => mFreeCount:            1154  --  LEAKED 242 !!!
}}

NB: I guess the leak is caused be interrupting the process, but reporting it just in case.

	Serge Gautherie (:sgautherie) Assignee
	Comment 10 • 16 years ago

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1a2pre) Gecko/2008080202 SeaMonkey/2.0a1pre] (home, debug default) (W2Ksp4)

Yes, there was some kind of "bad timing (/race)" condition between the functions (= threads).
Typical example (but not the only one) would be:
the writer sends 16 + 3, and registers 19,
the reader reads 16, registers 16, reads 3,
and |WaitForReceipt()| already returns with 16,
thus the assertion.

Core issue:
*Actually wait for all the (sent) data to arrive, instead of "some" only.
 *NB: |NS_ASSERTION(mReceived >= 0, ...);| used to be useless, as it was always true :-<

Unrelated:
*Merge 3 |printf()| into 1.
*Move a |buf[count] = '\0';| inside an |if|.
*Remove a useless |nsAutoCMonitor mon(this);|.

	Robert Strong (they/them - no direct email)
	Comment 11 • 16 years ago

Ran into this today... for what it is worth the patch looks good and fixes this bug.

	Brian Crowder
	Comment 12 • 16 years ago

Comment on attachment 332052 [details] [diff] [review]
(Av1) Fix broken logic ++
[Checkin: Comment 15]

This patch also fixes "make check" on my windows debug build (for this test, at least).

	Robert Strong (they/them - no direct email)
	Comment 13 • 16 years ago

Brian, are there other tests failing with make check on debug builds? If so, could you file them and make bug 448802 depend on them?

	Mike Shaver (:shaver -- probably not reading bugmail closely)
	Comment 14 • 16 years ago

Comment on attachment 332052 [details] [diff] [review]
(Av1) Fix broken logic ++
[Checkin: Comment 15]

r=shaver

	Robert Strong (they/them - no direct email)
	Comment 15 • 16 years ago

Pushed to mozilla-central

http://hg.mozilla.org/mozilla-central/index.cgi/rev/82033131d22fcfcf911a90a6bbdacb6b811df907