| Summary: | ICE(todt.c) struct literal initializing zero length array | ||
|---|---|---|---|
| Product: | D | Reporter: | Iain Buclaw <ibuclaw> |
| Component: | dmd | Assignee: | Walter Bright <bugzilla> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | bearophile_hugs, clugdbug, ibuclaw |
| Priority: | P2 | Keywords: | accepts-invalid, ice-on-invalid-code, patch |
| Version: | D2 | ||
| Hardware: | All | ||
| OS: | All | ||
| Attachments: |
PATCH against rev 755: fix crash in first case
Patch for 5090 |
||
Test cases:
struct A { int[0] b; }
A a = A(0); // Fails, compiler aborts
A b = {b:0}; // OK, but perhaps shouldn't be.
A c = A([]); // OK
A d = {b:[]}; // OK
Although the compiler shouldn't issue an assert, a zero length array
should be enforced to have a 0 length initializer.
Regards
This needs to be an error, because that 0 value has nowhere to be stored into:
struct A { int[0] b; }
A b = {b : 0};
Created attachment 813 [details]
PATCH against rev 755: fix crash in first case
Fixed:
A a = A(0); // Fails, compiler aborts
by issuing error msg.
I'll dig a bit more into fixing the second case, that should be an error as way.
Just under the "duplicate union" error is where to look iirc.
Created attachment 844 [details]
Patch for 5090
Reassigned to Walter. Attaching patch that turns assert into an error, but enforces that this corner case only happens for zero-length'd static arrays.
Make any improvements you may wish to add/remove, as I generally suck at coming up with good, concise error messages. :~)
Just to make note, these forms of initialisation are still OK for zero-length static arrays:
struct A { int[0] b; }
A a = A(); // OK
A b = {}; // OK
A c = A([]); // OK
A d = {b:[]}; // OK
Regards