Metadata 27+ Xref crash fix #110
Conversation
|
I think the pointers where just polluted so I couldn't just null check, but I'm not sure anymore |
|
This is still too brittle... I'm getting crashes with some functions again. |
|
Did anyone get a chance to look into this PR or the issue at hand? |
|
Any updates on this? @HookedBehemoth |
|
Works but the maintainers here don't seem to be interested in having this upstream |
In some version, I am getting Op0Kind = Register |
|
Do you actually know what this patch is for or do you just use this issue to report a bug that has nothing to do with this? |
I am not entirely sure if is related. But, using this PR or the original with Metadata 29, I get Argument out of range on certain xref. The op0kind is register. I supposed the address decoder has some issue. |
|
So it probably has nothing to do with this patch? |
On previous versions, one call per method instantiated all globals used. Now there is one call per global. Instead of a method-handle, the pointers are passed directly. With method inlining, multiple of these blocks are added into one method TODO: Only one of the flags is set and we might check the wrong one
HookedBehemoth
force-pushed
the
metadata27+fixes
branch
from
5db6bc6 to
dd30555
Compare
On previous versions, one call per method instantiated all globals used.
Now there is one call per global. Instead of a method-handle, the pointers are passed directly.
With method inlining, multiple of these blocks are added into one method
This bug would manifest as hard crashes during xref on methods that weren't fully initialized by the runtime.
TODO: Only one of the flags is set and we might check the wrong one
I got distracted and never properly implemented this. I hope these patches can help you.