C expression result-type model: ! width, shift result type, hex-constant typing

• Type: bug (C type semantics). Track C (C→IR lowering type computation).
• Found: 2026-07-06 c-testsuite run.

Failing tests

• 00178: sizeof(!a) prints 8, must be 4 — ! yields int, we type it as operand/64-bit.
• 00104: x = ~x; if (x != 0xffffffff) fails — int32 -1 vs 0xffffffff (unsigned int per C constant-typing rules) must compare equal after usual arithmetic conversions; we compare as 64-bit (-1 vs 4294967295).
• 00200: ISO C99 6.5.7p3 — X << T result type = promoted LEFT operand (battery checks sign+size of shift results via sizeof((M)+0) trick). Exit 1.

Root cause family

cfront types intermediate results too wide (64-bit) / ignores the C constant-type ladder (decimal vs hex, int → unsigned int → long ...) and integer-promotion result types for !, ~, <<, comparisons.

Gate

Drop 00104.c/00178.c/00200.c from test/c-conformance/pxx.skip; runner green. Sweep note: same model feeds sizeof, so check sizeof-of-expression paths after fix.

Update 2026-07-07

00178 (sizeof of a general non-ident expression) FIXED (commit 827cddfc). Remaining:

• 00104: ~x == 0xffffffff — C types a hex constant that overflows int as unsigned int; usual arithmetic conversions then make the int/unsigned compare equal. pxx types 0xffffffff as int64 -> unequal. Needs the C integer-constant type ladder (dec vs hex, int->unsigned->long) + conversion rules.
• 00200: X << T result type = promoted LEFT operand (a macro battery probing sign/size via sizeof((M)+0)).
• Also: an ident-starting sizeof expr sizeof(a<b) still sizes by the first ident (char=1) not the int result; same integer-type-model work.

Log

• 2026-07-08 — resolved, commit d513ca8f.