fix reinterpret(Char, ::UInt32) for "unnatural" values (fix #29181) (#…

…29192)

This code was assuming that character values only have bit-patterns
that decoding a string can produce, but of course `reinterpret` can
produce any bit pattern in a `Char` whatsoever. The fix doesn't use
that assumption and only uses the cache for actual ASCII characters.

src/datatype.c

@@ -689,8 +689,9 @@ static jl_value_t *boxed_char_cache[128];
 JL_DLLEXPORT jl_value_t *jl_box_char(uint32_t x)
 {
  jl_ptls_t ptls = jl_get_ptls_states();
- if (0 < (int32_t)x)
- return boxed_char_cache[x >> 24];
+ uint32_t u = bswap_32(x);
+ if (u < 128)
+ return boxed_char_cache[(uint8_t)u];
  jl_value_t *v = jl_gc_alloc(ptls, sizeof(void*), jl_char_type);
  *(uint32_t*)jl_data_ptr(v) = x;
  return v;

test/char.jl

@@ -271,11 +271,18 @@ end
  @test ncodeunits('\U10ffff') == 4
  # invalid encodings
  @test ncodeunits(reinterpret(Char, 0x80_00_00_00)) == 1
- @test ncodeunits(reinterpret(Char, 0x81_00_00_00)) == 1
- @test ncodeunits(reinterpret(Char, 0x80_80_00_00)) == 2
- @test ncodeunits(reinterpret(Char, 0x80_01_00_00)) == 2
- @test ncodeunits(reinterpret(Char, 0x80_00_80_00)) == 3
- @test ncodeunits(reinterpret(Char, 0x80_00_01_00)) == 3
- @test ncodeunits(reinterpret(Char, 0x80_00_00_80)) == 4
- @test ncodeunits(reinterpret(Char, 0x80_00_00_01)) == 4
+ @test ncodeunits(reinterpret(Char, 0x01_00_00_00)) == 1
+ @test ncodeunits(reinterpret(Char, 0x00_80_00_00)) == 2
+ @test ncodeunits(reinterpret(Char, 0x00_01_00_00)) == 2
+ @test ncodeunits(reinterpret(Char, 0x00_00_80_00)) == 3
+ @test ncodeunits(reinterpret(Char, 0x00_00_01_00)) == 3
+ @test ncodeunits(reinterpret(Char, 0x00_00_00_80)) == 4
+ @test ncodeunits(reinterpret(Char, 0x00_00_00_01)) == 4
+end
+
+@testset "reinterpret(Char, ::UInt32)" begin
+ for s = 0:31
+ u = one(UInt32) << s
+ @test reinterpret(UInt32, reinterpret(Char, u)) === u
+ end
 end