User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9pre) Gecko/2008042705 Minefield/3.0pre Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9pre) Gecko/2008042705 Minefield/3.0pre Run "dumpbin.exe /headers firefox.exe | grep checksum" (dumpbin ships with visual C++ in visual studio, but you can use any PE dumping tool) on a minefield win32 nightly and you'll see "0 checksum". This means that if you list modules with symbol problems in windbg (use windbg command "lm e") you'll see the firefox module listed: 0:003> lm e start end module name 00400000 00417000 firefox C (private pdb symbols) G:\symbols_cache\firefox.pdb\8ACD8EF1DE6140CDA4FF5E4C585358A32\firefox.pdb 71710000 71794000 COMCTL32 (export symbols) G:\symbols_cache\comctl32.dbg\44EF137484000\comctl32.dbg Also, note that the firefox module is marked with the "C" flag here indicating that the checksum information for this module is missing. Since windbg has pretty good heuristics it finds the symbols anyway but I bet some other tools need to have it available. Reproducible: Always Steps to Reproduce: 1. 2. 3.

We're not doing anything unusual, AFAIK, so if you have any ideas why this happens I'm all ears.

What is this checksum checksumming? Is it the contents of the PDB file, or the contents of the PE image itself, or something else?

It's a checksum of the binary, I believe: http://msdn.microsoft.com/en-us/library/ms680339%28v=VS.85%29.aspx http://www.woodmann.com/collaborative/tools/index.php/CheckSum_Fixer http://www.codeproject.com/KB/cpp/PEChecksum.aspx Seems kind of lame that we're using Microsoft's compiler and it doesn't bother to fill in this field.

Apparently you're supposed to pass the /RELEASE option to the linker to get this filled out: http://msdn.microsoft.com/en-us/library/vstudio/h8ksa72a.aspx

This works. I don't think the checksum actually does anything useful, but it stops WinDBG from complaining about it. This doesn't fix NSPR/NSS, but it fixes everything else.

This also disables incremental linking, which is not great! I filed bug 906119 about that.