| Message ID | 21f316ab-714a-58f6-a8d2-466d738b4ed3@web.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | test-ctype: test all classifiers | expand |
René Scharfe <l.s.r@web.de> writes: > Test the character classifier added by c2e9364a06 (cleanup: add > isascii(), 2009-03-07). It returns 1 for NUL as well, which requires > special treatment, as our string-based tester can't find it with > strcmp(3). Allow NUL to be given as the first character in a class > specification string. This has the downside of no longer supporting > the empty string, but that's OK since we are not interested in testing > character classes with no members. I wonder how effective a test we can have by checking a table we use in production (i.e. ctype.c::sane_ctype[]) against another table we use only for testing (i.e. string literals in test-ctype.c), but that is not something new in this series. I do not offhand know if the string literal prefixed with NUL is safe against clever compilers; my gut feeling says it should (i.e. allowing such an "optimization" does not seem to have much merit), but my gut has been wrong many times in this area, so... > > Signed-off-by: René Scharfe <l.s.r@web.de> > --- > t/helper/test-ctype.c | 19 +++++++++++++++++-- > 1 file changed, 17 insertions(+), 2 deletions(-) > > diff --git a/t/helper/test-ctype.c b/t/helper/test-ctype.c > index 92c4c2313e..caf586649f 100644 > --- a/t/helper/test-ctype.c > +++ b/t/helper/test-ctype.c > @@ -11,9 +11,14 @@ static void report_error(const char *class, int ch) > > static int is_in(const char *s, int ch) > { > - /* We can't find NUL using strchr. It's classless anyway. */ > + /* > + * We can't find NUL using strchr. Accept it as the first > + * character in the spec -- there are no empty classes. > + */ > if (ch == '\0') > - return 0; > + return ch == *s; > + if (*s == '\0') > + s++; > return !!strchr(s, ch); > } > > @@ -28,6 +33,15 @@ static int is_in(const char *s, int ch) > #define DIGIT "0123456789" > #define LOWER "abcdefghijklmnopqrstuvwxyz" > #define UPPER "ABCDEFGHIJKLMNOPQRSTUVWXYZ" > +#define ASCII \ > + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f" \ > + "\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f" \ > + "\x20\x21\x22\x23\x24\x25\x26\x27\x28\x29\x2a\x2b\x2c\x2d\x2e\x2f" \ > + "\x30\x31\x32\x33\x34\x35\x36\x37\x38\x39\x3a\x3b\x3c\x3d\x3e\x3f" \ > + "\x40\x41\x42\x43\x44\x45\x46\x47\x48\x49\x4a\x4b\x4c\x4d\x4e\x4f" \ > + "\x50\x51\x52\x53\x54\x55\x56\x57\x58\x59\x5a\x5b\x5c\x5d\x5e\x5f" \ > + "\x60\x61\x62\x63\x64\x65\x66\x67\x68\x69\x6a\x6b\x6c\x6d\x6e\x6f" \ > + "\x70\x71\x72\x73\x74\x75\x76\x77\x78\x79\x7a\x7b\x7c\x7d\x7e\x7f" > > int cmd__ctype(int argc, const char **argv) > { > @@ -38,6 +52,7 @@ int cmd__ctype(int argc, const char **argv) > TEST_CLASS(is_glob_special, "*?[\\"); > TEST_CLASS(is_regex_special, "$()*+.?[\\^{|"); > TEST_CLASS(is_pathspec_magic, "!\"#%&',-/:;<=>@_`~"); > + TEST_CLASS(isascii, ASCII); > > return rc; > } > -- > 2.39.1
Am 11.02.23 um 20:48 schrieb Junio C Hamano: > René Scharfe <l.s.r@web.de> writes: > >> Test the character classifier added by c2e9364a06 (cleanup: add >> isascii(), 2009-03-07). It returns 1 for NUL as well, which requires >> special treatment, as our string-based tester can't find it with >> strcmp(3). Allow NUL to be given as the first character in a class >> specification string. This has the downside of no longer supporting >> the empty string, but that's OK since we are not interested in testing >> character classes with no members. > > I wonder how effective a test we can have by checking a table we use > in production (i.e. ctype.c::sane_ctype[]) against another table we > use only for testing (i.e. string literals in test-ctype.c), but > that is not something new in this series. What aspect is left uncovered? Or do you mean that the production table should be made trivially readable to avoid having to test at all? I on the other hand wonder if we really should add more and more locale-ignoring classifiers. Parsing object headers and such sure require that stability, but parsing commit messages and blob payloads should perhaps better be done with locale-aware versions with multi-byte character support. > I do not offhand know if the string literal prefixed with NUL is > safe against clever compilers; my gut feeling says it should > (i.e. allowing such an "optimization" does not seem to have much > merit), but my gut has been wrong many times in this area, so... Some compilers do despicable things in the name of optimization, but I don't see the basis for truncating a string literal at the first NUL. C99 standard section 6.4.5 (String literals) paragraph 5 has a footnote that says: "A character string literal need not be a string (see 7.1.1), because a null character may be embedded in it by a \0 escape sequence." and 7.1.1 defines: "A string is a contiguous sequence of characters terminated by and including the first null character." René
René Scharfe <l.s.r@web.de> writes: > Am 11.02.23 um 20:48 schrieb Junio C Hamano: >> René Scharfe <l.s.r@web.de> writes: >> >>> Test the character classifier added by c2e9364a06 (cleanup: add >>> isascii(), 2009-03-07). It returns 1 for NUL as well, which requires >>> special treatment, as our string-based tester can't find it with >>> strcmp(3). Allow NUL to be given as the first character in a class >>> specification string. This has the downside of no longer supporting >>> the empty string, but that's OK since we are not interested in testing >>> character classes with no members. >> >> I wonder how effective a test we can have by checking a table we use >> in production (i.e. ctype.c::sane_ctype[]) against another table we >> use only for testing (i.e. string literals in test-ctype.c), but >> that is not something new in this series. > > What aspect is left uncovered? > > Or do you mean that the production table should be made trivially > readable to avoid having to test at all? Much closer to the latter but not quite. Both tables are not all that transparent, and it feels that the protection by the tests largely depends on the fact that it is less likely for us to make the same mistake in two "not so crystal clear" tables for the same byte. > ... but parsing commit messages and blob > payloads should perhaps better be done with locale-aware versions > with multi-byte character support. Yes, that does make sense but it is orthogonal to what sane_ctype wants to address, I would think.
Am 13.02.23 um 04:39 schrieb Junio C Hamano: > René Scharfe <l.s.r@web.de> writes: > >> Am 11.02.23 um 20:48 schrieb Junio C Hamano: >>> René Scharfe <l.s.r@web.de> writes: >>> >>>> Test the character classifier added by c2e9364a06 (cleanup: add >>>> isascii(), 2009-03-07). It returns 1 for NUL as well, which requires >>>> special treatment, as our string-based tester can't find it with >>>> strcmp(3). Allow NUL to be given as the first character in a class >>>> specification string. This has the downside of no longer supporting >>>> the empty string, but that's OK since we are not interested in testing >>>> character classes with no members. >>> >>> I wonder how effective a test we can have by checking a table we use >>> in production (i.e. ctype.c::sane_ctype[]) against another table we >>> use only for testing (i.e. string literals in test-ctype.c), but >>> that is not something new in this series. >> >> What aspect is left uncovered? >> >> Or do you mean that the production table should be made trivially >> readable to avoid having to test at all? > > Much closer to the latter but not quite. > > Both tables are not all that transparent, and it feels that the > protection by the tests largely depends on the fact that it is less > likely for us to make the same mistake in two "not so crystal clear" > tables for the same byte. The test strings for islower() and isupper() I wrote down from memory long ago, I think. They should be easily verifiable, like the new ones for isxdigit(). The one for isascii() is a bit tiring, but verifying it against the man page which says that the characters from value 0 to 0177 are included should be feasible. The ones for iscntrl() and ispunct() I got from their man page. But when it came to isprint() I got lazy and just copied from ctype.c -- you got me there. A more intuitive representation could be: " " LOWER UPPER DIGIT PUNCT In my experience having two copies already helps when modifying one of them -- but at least at some point we better check them against an external source of truth. The ctype.c version needs to be fast, so we probably have to make some concessions to readability. I'd love to be proven wrong on that, though. >> ... but parsing commit messages and blob >> payloads should perhaps better be done with locale-aware versions >> with multi-byte character support. > > Yes, that does make sense but it is orthogonal to what sane_ctype > wants to address, I would think. Currently we can only use one or the other variant because our sane versions use the same names as the locale-aware ones. Full overlap instead of orthogonality. I don't know if there is a practical impact besides not recognizing function lines that start with umlauts etc. for diff hunk headers, though. René
René Scharfe <l.s.r@web.de> writes: >> Yes, that does make sense but it is orthogonal to what sane_ctype >> wants to address, I would think. > > Currently we can only use one or the other variant because our sane > versions use the same names as the locale-aware ones. Full overlap > instead of orthogonality. Ah, that is true but slightly complicated. As long as the caller of walks the string byte-by-byte and calls ispunct() in each iteration, I do not think any "locale aware ispunct()" do much good to us. Once callers are made locale aware, I am not sure they would still want to call the isfoo() functions that have been know to be very much byte-oriented. Thanks.
diff --git a/t/helper/test-ctype.c b/t/helper/test-ctype.c index 92c4c2313e..caf586649f 100644 --- a/t/helper/test-ctype.c +++ b/t/helper/test-ctype.c @@ -11,9 +11,14 @@ static void report_error(const char *class, int ch) static int is_in(const char *s, int ch) { - /* We can't find NUL using strchr. It's classless anyway. */ + /* + * We can't find NUL using strchr. Accept it as the first + * character in the spec -- there are no empty classes. + */ if (ch == '\0') - return 0; + return ch == *s; + if (*s == '\0') + s++; return !!strchr(s, ch); } @@ -28,6 +33,15 @@ static int is_in(const char *s, int ch) #define DIGIT "0123456789" #define LOWER "abcdefghijklmnopqrstuvwxyz" #define UPPER "ABCDEFGHIJKLMNOPQRSTUVWXYZ" +#define ASCII \ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f" \ + "\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f" \ + "\x20\x21\x22\x23\x24\x25\x26\x27\x28\x29\x2a\x2b\x2c\x2d\x2e\x2f" \ + "\x30\x31\x32\x33\x34\x35\x36\x37\x38\x39\x3a\x3b\x3c\x3d\x3e\x3f" \ + "\x40\x41\x42\x43\x44\x45\x46\x47\x48\x49\x4a\x4b\x4c\x4d\x4e\x4f" \ + "\x50\x51\x52\x53\x54\x55\x56\x57\x58\x59\x5a\x5b\x5c\x5d\x5e\x5f" \ + "\x60\x61\x62\x63\x64\x65\x66\x67\x68\x69\x6a\x6b\x6c\x6d\x6e\x6f" \ + "\x70\x71\x72\x73\x74\x75\x76\x77\x78\x79\x7a\x7b\x7c\x7d\x7e\x7f" int cmd__ctype(int argc, const char **argv) { @@ -38,6 +52,7 @@ int cmd__ctype(int argc, const char **argv) TEST_CLASS(is_glob_special, "*?[\\"); TEST_CLASS(is_regex_special, "$()*+.?[\\^{|"); TEST_CLASS(is_pathspec_magic, "!\"#%&',-/:;<=>@_`~"); + TEST_CLASS(isascii, ASCII); return rc; }
Test the character classifier added by c2e9364a06 (cleanup: add isascii(), 2009-03-07). It returns 1 for NUL as well, which requires special treatment, as our string-based tester can't find it with strcmp(3). Allow NUL to be given as the first character in a class specification string. This has the downside of no longer supporting the empty string, but that's OK since we are not interested in testing character classes with no members. Signed-off-by: René Scharfe <l.s.r@web.de> --- t/helper/test-ctype.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) -- 2.39.1