[v4,2/2] preprocessor/106426: Treat u8 character literals as unsigned in char8_t modes.
Commit Message
This patch corrects handling of UTF-8 character literals in preprocessing
directives so that they are treated as unsigned types in char8_t enabled
C++ modes (C++17 with -fchar8_t or C++20 without -fno-char8_t). Previously,
UTF-8 character literals were always treated as having the same type as
ordinary character literals (signed or unsigned dependent on target or use
of the -fsigned-char or -funsigned char options).
PR preprocessor/106426
gcc/c-family/ChangeLog:
* c-opts.cc (c_common_post_options): Assign cpp_opts->unsigned_utf8char
subject to -fchar8_t, -fsigned-char, and/or -funsigned-char.
gcc/testsuite/ChangeLog:
* g++.dg/ext/char8_t-char-literal-1.C: Check signedness of u8 literals.
* g++.dg/ext/char8_t-char-literal-2.C: Check signedness of u8 literals.
libcpp/ChangeLog:
* charset.cc (narrow_str_to_charconst): Set signedness of CPP_UTF8CHAR
literals based on unsigned_utf8char.
* include/cpplib.h (cpp_options): Add unsigned_utf8char.
* init.cc (cpp_create_reader): Initialize unsigned_utf8char.
---
gcc/c-family/c-opts.cc | 1 +
gcc/testsuite/g++.dg/ext/char8_t-char-literal-1.C | 6 +++++-
gcc/testsuite/g++.dg/ext/char8_t-char-literal-2.C | 4 ++++
libcpp/charset.cc | 4 ++--
libcpp/include/cpplib.h | 4 ++--
libcpp/init.cc | 1 +
6 files changed, 15 insertions(+), 5 deletions(-)
Comments
On Tue, 2 Aug 2022, Tom Honermann via Gcc-patches wrote:
> This patch corrects handling of UTF-8 character literals in preprocessing
> directives so that they are treated as unsigned types in char8_t enabled
> C++ modes (C++17 with -fchar8_t or C++20 without -fno-char8_t). Previously,
> UTF-8 character literals were always treated as having the same type as
> ordinary character literals (signed or unsigned dependent on target or use
> of the -fsigned-char or -funsigned char options).
OK in the absence of C++ maintainer objections within 72 hours. (This is
the case where, when I added support for such literals for C (commit
7c5890cc0a0ecea0e88cc39e9fba6385fb579e61), I raised the question of
whether they should be unsigned in the preprocessor for C++ as well.)
On 8/2/22 6:14 PM, Joseph Myers wrote:
> On Tue, 2 Aug 2022, Tom Honermann via Gcc-patches wrote:
>
>> This patch corrects handling of UTF-8 character literals in preprocessing
>> directives so that they are treated as unsigned types in char8_t enabled
>> C++ modes (C++17 with -fchar8_t or C++20 without -fno-char8_t). Previously,
>> UTF-8 character literals were always treated as having the same type as
>> ordinary character literals (signed or unsigned dependent on target or use
>> of the -fsigned-char or -funsigned char options).
> OK in the absence of C++ maintainer objections within 72 hours. (This is
> the case where, when I added support for such literals for C (commit
> 7c5890cc0a0ecea0e88cc39e9fba6385fb579e61), I raised the question of
> whether they should be unsigned in the preprocessor for C++ as well.)
Joseph, would you be so kind as to commit this patch series for me? I
don't have commit access. Thank you in advance!
Tom.
On Mon, 8 Aug 2022, Tom Honermann via Gcc-patches wrote:
> On 8/2/22 6:14 PM, Joseph Myers wrote:
> > On Tue, 2 Aug 2022, Tom Honermann via Gcc-patches wrote:
> >
> > > This patch corrects handling of UTF-8 character literals in preprocessing
> > > directives so that they are treated as unsigned types in char8_t enabled
> > > C++ modes (C++17 with -fchar8_t or C++20 without -fno-char8_t).
> > > Previously,
> > > UTF-8 character literals were always treated as having the same type as
> > > ordinary character literals (signed or unsigned dependent on target or use
> > > of the -fsigned-char or -funsigned char options).
> > OK in the absence of C++ maintainer objections within 72 hours. (This is
> > the case where, when I added support for such literals for C (commit
> > 7c5890cc0a0ecea0e88cc39e9fba6385fb579e61), I raised the question of
> > whether they should be unsigned in the preprocessor for C++ as well.)
>
> Joseph, would you be so kind as to commit this patch series for me? I don't
> have commit access. Thank you in advance!
Done.
@@ -1062,6 +1062,7 @@ c_common_post_options (const char **pfilename)
/* char8_t support is implicitly enabled in C++20 and C2X. */
if (flag_char8_t == -1)
flag_char8_t = (cxx_dialect >= cxx20) || flag_isoc2x;
+ cpp_opts->unsigned_utf8char = flag_char8_t ? 1 : cpp_opts->unsigned_char;
if (flag_extern_tls_init)
{
@@ -1,6 +1,6 @@
// Test that UTF-8 character literals have type char if -fchar8_t is not enabled.
// { dg-do compile }
-// { dg-options "-std=c++17 -fno-char8_t" }
+// { dg-options "-std=c++17 -fsigned-char -fno-char8_t" }
template<typename T1, typename T2>
struct is_same
@@ -10,3 +10,7 @@ template<typename T>
{ static const bool value = true; };
static_assert(is_same<decltype(u8'x'), char>::value, "Error");
+
+#if u8'\0' - 1 > 0
+#error "UTF-8 character literals not signed in preprocessor"
+#endif
@@ -10,3 +10,7 @@ template<typename T>
{ static const bool value = true; };
static_assert(is_same<decltype(u8'x'), char8_t>::value, "Error");
+
+#if u8'\0' - 1 < 0
+#error "UTF-8 character literals not unsigned in preprocessor"
+#endif
@@ -1960,8 +1960,8 @@ narrow_str_to_charconst (cpp_reader *pfile, cpp_string str,
/* Multichar constants are of type int and therefore signed. */
if (i > 1)
unsigned_p = 0;
- else if (type == CPP_UTF8CHAR && !CPP_OPTION (pfile, cplusplus))
- unsigned_p = 1;
+ else if (type == CPP_UTF8CHAR)
+ unsigned_p = CPP_OPTION (pfile, unsigned_utf8char);
else
unsigned_p = CPP_OPTION (pfile, unsigned_char);
@@ -581,8 +581,8 @@ struct cpp_options
ints and target wide characters, respectively. */
size_t precision, char_precision, int_precision, wchar_precision;
- /* True means chars (wide chars) are unsigned. */
- bool unsigned_char, unsigned_wchar;
+ /* True means chars (wide chars, UTF-8 chars) are unsigned. */
+ bool unsigned_char, unsigned_wchar, unsigned_utf8char;
/* True if the most significant byte in a word has the lowest
address in memory. */
@@ -231,6 +231,7 @@ cpp_create_reader (enum c_lang lang, cpp_hash_table *table,
CPP_OPTION (pfile, int_precision) = CHAR_BIT * sizeof (int);
CPP_OPTION (pfile, unsigned_char) = 0;
CPP_OPTION (pfile, unsigned_wchar) = 1;
+ CPP_OPTION (pfile, unsigned_utf8char) = 1;
CPP_OPTION (pfile, bytes_big_endian) = 1; /* does not matter */
/* Default to no charset conversion. */