Philip Craig

Enable an efficent way of parsing only the compile unit DIE

1

In order to parse the compile unit DIE, we currently need to parse all abbreviations, and allocate storage for them. It would be nice to have a way of parsing...

See #451

For example, we should have a place to test the `LoongArchBranch` encoding that was added in #450

read/macho doesn't handle scattered relocations

1

Not sure how common these are. They aren't used for x86-64. We should at least stop objcopy from trying to copy them though.

write: redesign relocation kinds

5

I think that relocation kinds are a bit too artificial, and we should do something simpler. Currently relocation kinds are a combination of a `RelocationKind` and a `RelocationEncoding`. I think...

read/pe should handle unaligned section addresses and sizes

1

The PE file format specifies that sections should be aligned, but Windows still loads them if they are not. We should duplicate the logic that Windows uses.

The PE file format specifies that weak symbols should have an auxiliary record: https://docs.microsoft.com/en-us/windows/win32/debug/pe-format#auxiliary-format-3-weak-externals Additionally, we don't emit undefined weak symbols, and I'm not sure what these should look like....

Don't use Vec::with_capacity with untrusted lengths

3

This is currently used for ELF decompression, and even though it is good for performance, it allows malformed files to cause allocation of large amounts of memory.

This should be able to detect the error in https://github.com/fastly/lucet/pull/271#issuecomment-524006064

Untested, but it appears that sections with a type of `SHT_GROUP` will be parsed as `SectionKind::Unknown`, and written out with a `sh_type` set to `SHT_PROGBITS`. Even with that fixed, we'll...