Add option to disable alignment check #1332
Conversation
…static-morse Miri: let machine hook dynamically decide about alignment checks This is needed for rust-lang/miri#1332.
|
Will miri still report actual loads from unaligned addresses? Either way this is great -- it's been frustrating to disable tests for miri because they were too thorough and checked that every possible alignment and size combination didn't result in a misaligned load or OOB read. |
That is the exact check this disables. I am confused by your question.
Except for #1074, I do not know of a case where Miri is too thorough. Alignment errors otherwise are UB and indicate a bug in the program. |
I believe that's the case I mean. Basically I have code that manually aligns a pointer using pointer arithmetic. I test this with a very large number of possible input alignments and lengths to ensure it's sane, but miri assumes that any manually-aligned pointer is still invalid. As a result, I can't run my exhaustive tests under miri (and only some of the other less thorough ones). I had thought this was for disabling miri complaints about that case -- I guess it's about all cases wrt alignment. That doesn't bother me that much, since I generally have have a wrapper around ptr::read which asserts the alignment is correct for stuff like this, but yeah, I must have misunderstood. |
No. That would be resolving #1332. This does not resolve that issue, but it offers a work-around (at the expense of possibly missing bugs). |
RalfJung
force-pushed
the
disable-alignment-check
branch
from
330c1c9
to
77c0e7e
Compare
RalfJung
force-pushed
the
disable-alignment-check
branch
from
77c0e7e
to
8e73db6
Compare
|
@bors r+ |
|
📌 Commit 8e73db6 has been approved by |
Add option to disable alignment check Requires rust-lang/rust#71101 Fixes #1326
|
@bors retry r+ |
|
📌 Commit f4a1544 has been approved by |
|
☀️ Test successful - checks-travis, status-appveyor |
Requires rust-lang/rust#71101
Fixes #1326