I finally got tired of doing "history | grep diff" etc, and partially automated the daily valgrind run; the resulting script is at http://kegel.com/wine/valgrind/valgrind-daily.sh
The results are now in directories named like http://kegel.com/wine/valgrind/logs-2007-12-06/ A summary of the new problems compared to last run are in files named like http://kegel.com/wine/valgrind/logs-2007-12-06-summary.txt And next to each individual result, there is a diff with the previous run, e.g. http://kegel.com/wine/valgrind/logs-2007-12-06/vg-oleaut32_varformat.txt http://kegel.com/wine/valgrind/logs-2007-12-06/vg-oleaut32_varformat-diff.tx...
To look at what changed since yesterday, start by looking at the day's summary file. In it, there's one line per test, followed by the changes in that test's results, if any. Lines that start with + indicate new problems (boo); lines that start with - indicate old problems that are now gone (yay!) Nearby lines that are identical except for one starting with + and one starting with - indicate a problem that moved, so ignore those.
For example, in today's summary, http://kegel.com/wine/valgrind/logs-2007-12-06-summary.txt the first test with changes is in hlink:
diff -u logs-2007-12-05/vg-hlink_hlink.txt logs-2007-12-06/vg-hlink_hlink.txt - Syscall param write(buf) points to uninitialised byte(s)
That means the hlink test has one fewer error. Yay!
The next test with changes is
diff -u logs-2007-12-05/vg-ole32_moniker.txt logs-2007-12-06/vg-ole32_moniker.txt + Syscall param write(buf) points to uninitialised byte(s) - Syscall param write(buf) points to uninitialised byte(s) - Conditional jump or move depends on uninitialised value(s)
The 'Syscall' error just moved, so that's not really a change. The 'Conditional' error went away, yay!
The next test with changes is
diff -u logs-2007-12-05/vg-oleaut32_varformat.txt logs-2007-12-06/vg-oleaut32_varformat.txt + Invalid read of size 2
That's a + with no matching -, so it's a new error, boo! Let's see if it's a real error, or just crap in the log. The first thing to check is the diff of that test's results with the last run: http://kegel.com/wine/valgrind/logs-2007-12-06/vg-oleaut32_varformat-diff.tx... It shows a fairly convincing new error:
+ Invalid read of size 2 + at VarFormatFromTokens (varformat.c:1888) + by VarFormat (varformat.c:2096) + by func_varformat (varformat.c:317) + by run_test (test.h:387) + by main (test.h:436) + Address 0x7F00AB01 is 15 bytes before a block of size 8 alloc'd + at RtlAllocateHeap (heap.c:1225) + by SysAllocStringByteLen (oleaut.c:357) + by VariantCopy (variant.c:722) + by VariantChangeTypeEx (variant.c:992) + by VarFormatFromTokens (varformat.c:1884) + by VarFormat (varformat.c:2096) + by func_varformat (varformat.c:317) + by run_test (test.h:387) + by main (test.h:436)
Now, sometimes tests are flaky, and errors just happen to come and go, so this doesn't mean for sure it's a new error. But looking back a few days, I don't see it before. So it's probably a new error. Let's see, who's been submitting changes to oleaut32 lately? Only change in git since the last run seems to be http://www.winehq.org/pipermail/wine-cvs/2007-December/038491.html It's not clear how that patch could have caused the error offhand, but it's worth emailing the author and asking him to check, so I've cc'd him.
And that's the only new error in the whole run.
It's worth watching for new errors and alerting the authors, since the change is fresh in their mind, they'll probably be able to fix it quickly. Egg in the face is a dish best served warm.
I'll try to keep running this script daily, but it'd be nice if somebody else took over (and possibly even increased the automation; I still have to do git pull, run the script, and upload by hand.) It doesn't help that valgrind only works well on some machines, but hey, what's life without challenges. - Dan
-
Dan Kegel