Support MIPS64 .o files - don't remove has_addend

[noamraph]

A slightly modified #493

[eliben]

Thanks, this looks better to me. @sevaa WDYT?

[eliben]

I like that this decouples the has_addend refactoring from the specific MIPS64 change. We can consider a separate has_addend refactoring PR once this one lands.

@sevaa specifically I'd like you to review the run_dwarfdump_tests changes here

[sevaa]

@noamraph: may I hear more about the (0x0000000000000000 ".text") discrepancy? Where is it, exactly?

In general, we prefer to make the output of our scripts match the reference tool output, rather than introduce special cases on testing. If the tool output format evolves over time (like readelf's) - oh well, life goes on.

[noamraph]

@sevaa here's the diff:

I tried modifying dwarfdump.py for just adding ".text" for DW_FORM_addr with value=0, but it turns out that in other ELF files llvm-dwarfdump didn't add it. I'm not sure why llvm-dwarfdump added the ".text" here and not in other files. It may be related to this being a .o.

I can modify dwarfdump.py to add ".text" to DW_FORM_addr with value=0 on "relocatable" ELF files. Would that be better than just ignoring it in the output?

[sevaa]

Have you checked the llvm-dwarfdump sources?

[noamraph]

@sevaa I tried to look in llvm-dwarfdump sources, and didn't find the place where it prints the address.
Is there a chance we could continue without going to the root of this? After all, the DWARF parsing is obviously correct. The dwarfdump output is also correct, it just doesn't match exactly the llvm-dwarfdump output. The question is, could we add support for MIPS64 .o DWARF parsing without understanding when exactly llvm-dwarfdump prints the section name.

Thanks,
Noam

[sevaa]

Every time we introduce a special case treatment in the test runner, it leaves a bad taste in my mouth :) That said, no hard veto.

[noamraph]

@sevaa I feel the bad taste as well, and I would really like to understand llvm-dwarfdump further and update dwarfdump.py accordingly. But there's a good chance I won't have the time. So I would understand it if you will reject adding support for MIPS64 .o in order to keep the test clean, but I would appreciate it if you would agree to add it.

Anyway, I pushed another commit which I think makes the special-case a bit more localized in run_dwarfdump_test.py, and adds a comment explaining it.

What do you think?

[eliben]

The test runner exception is now targeted enough, I feel alright with it. Merging.