| commit | 795400d899d414ce52e552d7bcd2e6476b4f0ba4 | |
| author | jdelong <jdelong@fb.com> | |
| Sun, 3 Feb 2013 23:18:10 +0000 (3 15:18 -0800) | ||
| committer | Sara Golemon <sgolemon@fb.com> | |
| Mon, 11 Feb 2013 11:44:07 +0000 (11 03:44 -0800) | ||
| tree | f4a3fa032d9993de7ee354164a7ff1ab4b1f3eff | treesnapshot (tar.gz zip) |
| parent | 5483751a255271598c5bb31eb46191fe84977bd9 | commitdiff |
Table-drive type checking for IR opcodes
The way type.cpp was handling return types wasn't quite
sufficient for multiple destinations and unfortunately added a step
that's easy to forget when you add new opcodes with the HasDest flag.
Also, assertions about operand types are too distributed through the
code. Pull everything into a central table, similar to hhbc.h. This
still leaves several opcodes without type signatures, since figuring
it out got a bit tedious for now. It did uncover a few documentation
bugs in ir.specification, though.
The way type.cpp was handling return types wasn't quite
sufficient for multiple destinations and unfortunately added a step
that's easy to forget when you add new opcodes with the HasDest flag.
Also, assertions about operand types are too distributed through the
code. Pull everything into a central table, similar to hhbc.h. This
still leaves several opcodes without type signatures, since figuring
it out got a bit tedious for now. It did uncover a few documentation
bugs in ir.specification, though.