I'll note that the requirement that the # call args and flags are i32s conflicts with the rest of the documentation, namely, that the flags argument is 64 bits wide.

I don't have commit access, so if you could merge this patch for me that would be great.

In D24073#539679, @rovka wrote:

In D24073#539438, @kavon wrote:

When the op code is a patchpoint, isn't a call still emitted? I think maybe the total number of bytes returned for patchpoints should include the bytes for the call.

From my reading of AArch64AsmPrinter::LowerPATCHPOINT, it looks like the number of bytes requested by the patchpoint has to include the bytes for the call - i.e. if the call is generated, its size is subtracted from the number of bytes requested, and only the remaining number of nops are added.

When the op code is a patchpoint, isn't a call still emitted? I think maybe the total number of bytes returned for patchpoints should include the bytes for the call.

Okay, here's a version with the stackmap version bumped to 2.

This revision includes the changes requested in the inline comments.

(added replies to inline comments ahead of adding a new revision)

In D23487#522475, @sanjoy wrote:

Hi @kavon,

In D23487#521887, @kavon wrote:

this looks like the kind of thing that would involve bumping the stackmap version from 1 to 2.

@sanjoy and others: if this is decided, then the patch will have to change quite a lot, so please let me know!

Sorry, I forgot about this bit in the last review. I think the version number needs to change. While I agree that this can be seen as "bugfix" kind of change, in practice changing the format without a version bump will silently break backends.

@ributzka The bug that is fixed by this patch helps all users of stackmaps (gc.statepoint and patchpoints), because you couldn't associate callsites with their functions. I think we should at least merge this fix in so that the current stackmap format is actually usable.

Updated per sanjoy's inline feedback.

this looks like the kind of thing that would involve bumping the stackmap version from 1 to 2.

I updated the original patch so that the record counts are 8 bytes wide instead of 4 bytes wide to fix an alignment issue in the StackMapParser.