Splited functions should have splited lines

[calixteman]

When a function is splited several FUNCs are emitted but only the last one has lines info.
But if an address is in the first part, since the first FUNC hasn't line info, then we haven't lines for it.
An example:
https://crash-stats.mozilla.com/report/index/f43a2011-b0f0-4881-bfe6-a471c0191024
Thread 19, frame (0 + 5)
If we put lines (in breakpad file) after "FUNC 33458 41 0 CRpcThreadCache::RpcWorkerThreadEntry(void *)" in combase.sym we get the correct line (63).
Or maybe it's a bug in stackwalker (at least when the sym file is parsed).

[gabrielesvelto]

This is an interesting issue! When I implemented support for split address ranges for Linux symbols (DW_AT_ranges DWARF attribute) I handled this by splitting the lines so that one would have:

FUNC 1000 10 0 foobar
1000 f 10 1
FUNC 2000 10 0 fobar
2000 f 11 1
etc...

It might be that the Windows version of dump_syms is not handling this correctly. The correct behaviour is to have multiple FUNC entries each one with the lines that belong to the range specified in FUNC.

I don't think we should try and fix this in the stack walker because there might be infinite combinations of FUNC and line entries and handling them could be brittle. Better make sure that our debuginfo is correct.

[luser]

FYI I had added support to the Breakpad dump_syms for this ages ago but I'm not sure how well concepts from the DIA SDK map to PDB internals.