Add "b" prefix when using DREQ_GET_MEM_BINARY

[khang06]

Fixes #3901.

GDB errors out if the response doesn't start with "b" (see https://github.com/bminor/binutils-gdb/blob/master/gdb/remote.c#L9739). This works on GDB 16.1 and LLDB 19.1.7, but I haven't tested anything else.

[khuey]

It appears older LLDB is unhappy with this.

[khang06]

It appears older LLDB is unhappy with this.

Hmmm, the failing tests also break on latest LLDB too. It seems like LLDB doesn't skip the first character even though the protocol specification says it's required. Not really sure what to do about that other than disabling binary memory read/write command support, since LLDB appears to have a fallback path for when that isn't supported.

[rocallahan]

So GDB implemented this in a completely incompatible way to LLDB?

[rocallahan]

Yes, OK, that's what they did.

@khang06 can you contact the GDB people or file a GDB bug and tell them that they've implemented this incompatibly with LLDB and now all third-party gdbservers that implemented this in an LLDB-compatible way are broken?

It would be nice if they could fix this by accepting output either with or without the 'b', but that doesn't really work because 'b' could be legitimate data :-(.

If we have to we can work around this in rr by having an "LLDB mode" flag and setting that flag automatically by detecting use of LLDB-specific extensions, but it would really suck for all already-installed versions of rr to not work with gdb 16.1. Seems better to issue a gdb 16.1 point release as soon as possible.

[rocallahan]

Yeah so I think we should ask the GDB folks if they are willing to change to be LLDB compatible here.
Whether or not they agree to that, I think we should add an LLDB mode flag to rr with a command-line option to control it, and require LLDB users to explicitly opt-in to the LLDB flavour protocol. We should disable support for LLDB's extension packets when that flag is not set. Then if GDB makes more incompatible extensions (or LLDB does) hopefully rr won't break.

[rocallahan]

Yeah so I think we should ask the GDB folks if they are willing to change to be LLDB compatible here.

I'm doing this.

[khang06]

Could you make a GDB 16.1 point release that removes the 'b'? AFAIK it serves no purpose.

I think that's there to differentiate successful cases from the "E" prefix error cases according to the protocol page.

[rocallahan]

See https://sourceware.org/pipermail/gdb/2025-January/051705.html

[rocallahan]

I think that's there to differentiate successful cases from the "E" prefix error cases according to the protocol page.

Good point, that makes sense.