| // RUN: %clang_cc1 -std=c23 %s --embed-dir=%S/Inputs -fsyntax-only -verify | |
| const char data[] = { | |
| #embed <single_byte.txt> prefix('\xA', ) | |
| }; | |
| const char empty_data[] = { | |
| #embed <media/empty> prefix('\xA', ) | |
| 1 | |
| }; | |
| static_assert(sizeof(data) == 2); | |
| static_assert('\xA' == data[0]); | |
| static_assert('b' == data[1]); | |
| static_assert(sizeof(empty_data) == 1); | |
| static_assert(1 == empty_data[0]); | |
| struct S { | |
| int x, y, z; | |
| }; | |
| const struct S s = { | |
| #embed <single_byte.txt> prefix( .x = 100, .y = 10, ) | |
| }; | |
| static_assert(s.x == 100); | |
| static_assert(s.y == 10); | |
| static_assert(s.z == 'b'); | |
| // Ensure that an empty file does not produce any prefix tokens. If it did, | |
| // there would be random tokens here that the parser would trip on. | |
| #embed <media/empty> prefix(0) | |
| // Ensure we diagnose duplicate parameters even if they're the same value. | |
| const unsigned char a[] = { | |
| #embed <jk.txt> prefix(1,) limit(1) prefix(1,) | |
| // expected-error@-1 {{cannot specify parameter 'prefix' twice in the same '#embed' directive}} | |
| , | |
| #embed <jk.txt> prefix(1,) if_empty() prefix(2,) | |
| // expected-error@-1 {{cannot specify parameter 'prefix' twice in the same '#embed' directive}} | |
| }; |