Paolo Bonzini wrote: > On 06/02/2011 11:08 PM, Jim Meyering wrote: >> #if MBS_SUPPORT >> - int b2 = wctob ((unsigned char) b); >> - if (b2 == EOF || b2 == b) >> + /* Below, note how when b2 != b and we have a uni-byte locale >> + (MB_CUR_MAX == 1), we set b = b2. I.e., in a uni-byte locale, >> + we can safely call setbit with a non-EOF value returned by wctob. >> */ >> + int b2 = wctob (b); >> + if (b2 == EOF || b2 == b || (MB_CUR_MAX == 1 ? (b=b2), 1 : 0)) > > Can you explain again the reason for testing "b2 == EOF"? It seems > wrong, and without it you can just make > > if (MB_CUR_MAX == 1 || b2 == b) > setbit ((unsigned char) b, c);
Hi Paolo, Your test would disable DFA-based matching for some bytes in a locale like ru_RU.KOI8-R, because a pattern like [\360] leads to "wint_t b" having the value 1055 (0x041F), and that is obviously too large to be used as the first argument to setbit. However, converting that "B" back to a single-byte value, B2, gives us back \360, which is ok to use there. Hence the "(b=b2)" part of that admittedly ugly expression. The b2 == EOF part is required for the somewhat similar bug I fixed a month ago: fix a bug whereby echo c|grep '[c]' would fail for any c in 0x80..0xff 8da41c930e03a8635cbd8c89e3e591374c232c89 The corresponding test demonstrates the need: tests: exercise bug with 0x80..0xff in [...] d98338ebf842ec9b69631837eee50ebdcd543505 Thanks for the feedback. If you see a better way, I'm sure you'll let me know. BTW, seeing your cast, I now think it'd be prudent to guard that setbit use: #if MBS_SUPPORT /* Below, note how when b2 != b and we have a uni-byte locale (MB_CUR_MAX == 1), we set b = b2. I.e., in a uni-byte locale, we can safely call setbit with a non-EOF value returned by wctob. */ int b2 = wctob (b); if (b2 == EOF || b2 == b || (MB_CUR_MAX == 1 ? (b=b2), 1 : 0)) #endif if (b < 256) setbit (b, c); } -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org