[REG2.067a] Issue 13999 - Associative array literal with static array keys must now have matching key lengths #4307
Conversation
9rnsr
changed the title
9rnsr
force-pushed
the
fix13999
branch
2 times, most recently
from
1f54053
to
f3e2a5d
Compare
| void test13907() | ||
| { | ||
| static assert(!__traits(compiles, { f13907_1("\U00010000"w); })); | ||
| static assert(!__traits(compiles, { f13907_1("\U00010000" ); })); | ||
| f13907_2("\U00010000"w); | ||
| f13907_2("\U00010000"); | ||
| static assert(!__traits(compiles, { f13907_3("\U00010000"w); })); | ||
| static assert(!__traits(compiles, { f13907_3("\U00010000" ); })); | ||
| //f13907_3("\U00010000"w); // Re-enable implicit length extension, from issue 13999 (unlisted codegen bug) |
There was a problem hiding this comment.
What are the details of the codegen bug?
There was a problem hiding this comment.
The line segfaults at runtime.
There was a problem hiding this comment.
How do you produce the segfault? What platform/args? I'm not seeing it on win32.
There was a problem hiding this comment.
It happens in win32 platform, without any args. Test case:
void f13907_3(wchar[3] a) {}
void main()
{
f13907_3("\U00010000"w);
}Note that, git-head fails to compile the code. To reproduce runtime segfault, you need to use 2.066 or cherry-pick this PR.
There was a problem hiding this comment.
The error appears to happen because the string literal has type wchar[3] but se->len is set to 2.
wchar[2] can be passed in a single 32-bit stack slot, while wchar[3] requires two. As the calling convention uses callee-cleanup, the mismatch in allocated stack space causes a crash when the caller allocates only 4 bytes, and the callee frees 8, then loads the return address from the wrong offset.
I can see a couple of ways to deal with this:
- Update len when an implicit cast to
T[N]is performed (I think something similar is done when character width changes?) - Change e2ir.c:1277 to pass the static array's dim to
toStringSymbolinstead ofse->len(there may be other places that need to be updated)
I think 1 is probably the safest option.
There was a problem hiding this comment.
OK, I found and fixed a bug in StringExp::castTo().
…st now have matching key lengths Re-enable implicit length extension, eg. "abc" to char[5]
|
Auto-merge toggled on |
[REG2.067a] Issue 13999 - Associative array literal with static array keys must now have matching key lengths
https://issues.dlang.org/show_bug.cgi?id=13999
Re-enable implicit length extension, eg.
"abc"tochar[5].