Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Port to NetBSD, where, e.g., 'time_t' is wider than 'long'.
I audited the code and fixed as many width-asssumptions as I could find, including several places where the code assumed that 'time_t' was no wider than 'long'; this assumption is not true on 32-bit NetBSD platforms. This caught every problem that is already fixed in the NetBSD zic.c, and caught quite a few more. * Makefile: Add comments re HAVE_DOS_FILE_NAMES and HAVE_INTTYPES_H. * date.c (checkfinal, netsettime): Don't use 'long' where 'int' will do. * difftime.c (difftime): Mark with ATTRIBUTE_CONST. Use uintmax_t, not unsigned long, for the widest unsigned integer type. Use long double, not double, if time_t is wider than uintmax_t; this can in theory help on nonstandard platforms, such as GCC with 64-bit uintmax_t and 128-bit __int128_t. * localtime.c (struct ttinfo.tt_gmtoff, struct rule.r_time) (detzcode, getsecs, getoffset, gmtsub, localsub, increment_overflow32) (normalize_overflow32, time1, time2, timesub, transtime, tzparse) (time2sub, timeoff, gtime): * tzfile.h (SECSPERDAY): * zdump.c (SECSPERDAY): * zic.c (convert, puttzcode): Use int_fast32_t, not long, when all we care is that values up to 2**31 can be stored. This doesn't fix any bugs, but it allows more opportunity for compiler optimization. (struct lsinfo.ls_corr, timesub, leapcorr): Use int_fast64_t, not long, when values up to 2**63 can be stored. (timesub): Make it clearer when we are truncating 0.5 to 0. (increment_overflow32): Rename from long_increment_overflow. All uses changed. (normalize_overflow32): Rename from long_normalize_overflow. All uses changed. * private.h (HAVE_INTTYPES_H, ATTRIBUTE_CONST): New macros. Include <inttypes.h> if HAVE_INTTYPES_H. (INT_FAST64_MIN, INT_FAST64_MAX, SCNdFAST64, int_fast32_t, PRIdMAX) (uintmax_t, PRIuMAX, _Noreturn): Define to reasonable values if it's an older compiler. * scheck.c (scheck): Add support for arbitrary formats, such as those that SCNdFAST64 can expand to, at the price of no longer supporting weird conversion specs like "%[%]". * strftime.c (_fmt): Use intmax_t and uintmax_t to format time_t, not long and unsigned long. * zdump.c (int_fast32_t, intmax_t, PRIdMAX, SCNdMAX): Define for pre-C99 compilers, like private.h does. (delta, yeartot, main): Use intmax_t, not long. (hunt): Use time_t, not long, since the diff must be nonnegative. (tformat): Allow for time_t wider than long. * zic.c (ZIC_MIN, ZIC_MAX, SCNdZIC): New macros. (OFFSET_STRLEN_MAXIMUM, RULE_STRLEN_MAXIMUM): Remove. (struct rule): Make r_loyear, r_hiyear, r_tod, r_stdoff, z_gmtoff, z_stdoff zic_t, not long. (addtype, gethms, oadd, rpytime, tadd, gmtoffs, corr, inleap) (stringoffset, stringrule, outzone, addtype, adjleap, rpytime) (LDAYSPERWEEK): Use zic_t, not long. (leapminyear, leapmaxyear, min_year, max_year, rulesub, updateminmax) (outzone, rpytime): Use zic_t, not int. (usage): Now _Noreturn. (main): Use S_IWGRP, not 'unix', to determine whether to call umask. (writezone): Omit unnecessary cast. (mkdirs): Use HAVE_DOS_FILE_NAMES, not 'unix', to determine whether to parse DOS file anmes. (eitol): Remove; no longer needed.
- Loading branch information