Home My Page Projects MPFR
Summary Activity Forums Tracker Lists Tasks Docs News SCM Files

Activity

Activity:
Start Date:
End Date:
Time Activity By
2017-05-24
17:18:58 Source Code scm commit: [vl-tests/mpfrtests.sh] Added a comment about the additional tests. (r11535) Vincent Lefèvre
15:50:00 Source Code scm commit: [vl-tests/mpfrtests.{data,sh}] Update. (r11534) Vincent Lefèvre
11:33:50 Source Code scm commit: [tests/tsprintf.c] Fixed setlocale usage (problem detected with MinGW) by not using the returned string any longer. Indeed, ISO C99 says that this string "may be overwritten by a subsequent call to the setlocale function". (r11533) Vincent Lefèvre
09:15:54 Source Code scm commit: [src] Moved "#include <errno.h>" from printf.c to vasprintf.c (this should have been done in r5230, where EOVERFLOW was moved). This fixes the "Got r = -1, errno = 0" errors in tprintf with MPFR_CHECK_LARGEMEM=1. (r11532) Vincent Lefèvre
09:09:53 Source Code scm commit: [src/vasprintf.c] Added overflow detection at the end (due to the switch to mpfr_intmax_t for the string length in the string_buffer structure, it was no longer detected; however, this bug was not normally visible in the tests because here, a negative value was returned thanks to wrapping in a type conversion and the negative value was changed to the expected -1 value in printf.c). (r11531) Vincent Lefèvre
08:47:36 Source Code scm commit: [src/vasprintf.c] Added log messages. (r11530) Vincent Lefèvre
08:30:55 Source Code scm commit: [src/vasprintf.c] In case of overflow, do not stop before everything has been processed, allowing the %...n arguments to be filled, as decided. (r11529) Vincent Lefèvre
08:12:56 Source Code scm commit: [tests/tprintf.c] check_long_string() error handling: initialize n2 to an invalid value and do all tests before exiting. (r11528) Vincent Lefèvre
08:01:28 Source Code scm commit: [src/vasprintf.c] Fixed some parts of overflow tracking. (r11527) Vincent Lefèvre
07:49:39 Source Code scm commit: [tests/tprintf.c] More detailed error message. (r11526) Vincent Lefèvre
07:46:26 Source Code scm commit: [tests/tprintf.c] Added tests with %ln. (r11525) Vincent Lefèvre
07:15:40 Source Code scm commit: [tests/tprintf.c] Improved check_vprintf_failure(). (r11524) Vincent Lefèvre
07:10:21 Source Code scm commit: [src/vasprintf.c] Fixed the handling of the 'n' format specifier for size = 0, still allowing its support in case of overflow on the return value (i.e. when the number of characters that would have been output is > INT_MAX). Note: full support in case of overflow is still limited by GMP and the C library. [tests/tprintf.c] Increased the memory needed in check_long_string(). (r11523) Vincent Lefèvre
2017-05-23
13:17:12 Source Code scm commit: [src/mpfr-intmax.h] Define mpfr_uintmax_t and MPFR_INTMAX_MAX too. However, this will not necessarily be used by src/vasprintf.c, as this can be regarded as an EOVERFLOW error for which the behavior on the %...n values is not specified, i.e. we may not need to support size values larger than INT_MAX. (r11522) Vincent Lefèvre
2017-05-22
15:07:36 Source Code scm commit: [src/vasprintf.c] In a macro definition, changed a parameter to lowercase so that it doesn't look like a constant to mpfrlint. (r11521) Vincent Lefèvre
14:08:07 Source Code scm commit: [src/vasprintf.c] Added a FIXME for the percent_n() / mixed() bug due to the 'n' format specifier with size = 0. The way to fix this issue depends on the specifications in case of overflow on the return value, and for POSIX, they are unclear. Waiting for comments... (r11520) Vincent Lefèvre
12:33:48 Source Code scm commit: [src/mpfr-intmax.h] Define mpfr_intmax_t as being the largest signed integer type available for the MPFR build. It will be useful for the formatted output functions (src/vasprintf.c) to handle the 'n' format specifier, assuming that the value may be larger than INT_MAX, which is possible assuming that the C standard has a defect in the corresponding specification: https://sourceware.org/bugzilla/show_bug.cgi?id=21360 (r11519) Vincent Lefèvre
11:38:10 Source Code scm commit: [tests/tsprintf.c] Added explanation for the bugs triggered by percent_n() and mixed(). (r11518) Vincent Lefèvre
11:31:16 Source Code scm commit: [tests/tsprintf.c] Added percent_n() test to reproduce the bug found in mixed(), test without random data. (r11517) Vincent Lefèvre
10:53:50 Source Code scm commit: [src/vasprintf.c] In partition_number(), complete check for overflow on total. (r11516) Vincent Lefèvre
09:29:51 Source Code scm commit: [src/vasprintf.c] Partially fixed check for integer overflow when the long and int types have the same size. (r11515) Vincent Lefèvre
09:23:57 Source Code scm commit: [src/vasprintf.c] With assertions enabled, check that the number of characters to be written computed in partition_number() corresponds to what is really written. (r11514) Vincent Lefèvre
09:17:51 Source Code scm commit: [src/vasprintf.c] Fixed the bug triggered by the da_DK locale test: the computation of the number of characters to be written was incorrect in partition_number() when the thousands separator was not empty. This bug was visible with size = 0 and could be visible near overflow on the number of characters. Note: the buggy code was introduced in r5288 (2008-02-15), which fixed other issues with the thousands separator. (r11513) Vincent Lefèvre
09:13:33 Source Code scm commit: [tests/tsprintf.c] Do more tests in the da_DK locale (these are the only tests for which the thousands separator is not empty). (r11512) Vincent Lefèvre
08:56:55 Source Code scm commit: [tests/tsprintf.c] Do more tests involving check_sprintf() in a loop (to test mpfr_snprintf with size = 0 more often), and call mixed() later in order to get failures without the use of %n first. (r11511) Vincent Lefèvre
2017-05-19
15:04:24 Source Code scm commit: get rid of computations with 'double' in mpfr_zeta (r11510) Paul Zimmermann
14:47:51 Source Code scm commit: [tests/tsprintf.c] Added a TODO about %n test. (r11509) Vincent Lefèvre
14:04:46 Source Code scm commit: [doc/README.dev] Added MPFR_TESTS_FPE_TRAP documentation. (r11508) Vincent Lefèvre
13:59:19 Source Code scm commit: [tests/tests.c] Corrected comment about MPFR_TESTS_FPE_TRAP. (r11507) Vincent Lefèvre
13:57:15 Source Code scm commit: Renamed MPFR_TESTS_DIVBYZERO to MPFR_TESTS_FPE_DIV (less confusing). In doc/README.dev, updated the corresponding description. (r11506) Vincent Lefèvre
13:49:50 Source Code scm commit: [tests] Renamed MPFR_RAISE_EXCEPTIONS to MPFR_TESTS_FPE_TRAP. (r11505) Vincent Lefèvre
13:45:10 Source Code scm commit: [tests] Got rid of MPFR_TESTS_EXCEPTIONS (added in r11489 and above), which is redundant with MPFR_TESTS_DIVBYZERO and MPFR_ERRDIVZERO. The code related to these exceptions is now like r11488 (see the output of "svn diff -r11488"), but keeping the following changes: * Missing !defined(MPFR_ERRDIVZERO) tests added in r11498,11499. * In tests.c: - code related to traps (macro MPFR_RAISE_EXCEPTIONS, which will be renamed); - more detailed output about which FP exceptions occurred; - removal of "#ifdef MPFR_ERRDIVZERO" (as in r11493), as this is better with Automake 1.13+ (required by MPFR since r9375). Note: EXCEPTIONS_FLAG renamed to FPE_FLAGS, since EXCEPTIONS_FLAG is reserved by ISO C when <errno.h> is included. (r11504) Vincent Lefèvre
11:54:29 Source Code scm commit: [tests/tsprintf.c] Do the locale_da_DK() test several times to yield a failure by default once the bug triggered by mixed() is fixed. (r11503) Vincent Lefèvre
11:24:03 Source Code scm commit: [tests/tsprintf.c] Updated comment about r11501: the failure occurs with %n when the size argument of mpfr_vsnprintf is 0. (r11502) Vincent Lefèvre
11:08:14 Source Code scm commit: [tests/tsprintf.c] Do more tests with %n. One now gets a failure by default (which occurs when one gets randsize <= 0). (r11501) Vincent Lefèvre
10:45:25 Source Code scm commit: [tests/tsprintf.c] Fixed portability issues: The va_end(3) man page of Linux Programmer's Manual says: "On some systems, va_end contains a closing '}' matching a '{' in va_start, so that both macros must occur in the same function, and in a way that allows this." (r11500) Vincent Lefèvre
09:26:18 Source Code scm commit: [tests/tget_set_d64.c] Disable tests with a division by 0 (which now occurs only with -DXDEBUG) if MPFR_ERRDIVZERO is defined. (r11499) Vincent Lefèvre
09:20:19 Source Code scm commit: [tests/tset_ld.c] Disable an overflow test with a division by 0 (now occurs only with -DXDEBUG) if MPFR_ERRDIVZERO is defined. (r11498) Vincent Lefèvre
08:54:32 Source Code scm commit: [doc/README.dev] Details for "-DMPFR_TESTS_DIVBYZERO -DMPFR_ERRDIVZERO". (r11497) Vincent Lefèvre
08:34:00 Source Code scm commit: [tests/tsprintf.c] In mixed(), initialize n2 to an invalid value in order to ease failure analysis. (r11496) Vincent Lefèvre
08:15:46 Source Code scm commit: removed leftover debug statement in last commit (r11495) Paul Zimmermann
08:14:58 Source Code scm commit: fix for clang bug (https://bugs.llvm.org//show_bug.cgi?id=17686) (r11494) Paul Zimmermann
08:04:05 Source Code scm commit: [tests/tests.c] Change for MPFR_TESTS_DIVBYZERO: in case of FE_DIVBYZERO or FE_INVALID, fail unconditionally, not just when MPFR_ERRDIVZERO is defined, otherwise one cannot see the output about the floating-point exceptions with "make check" (the test of MPFR_ERRDIVZERO here should have been removed after switching to Automake 1.13, which discards the output when the test program succeeds). Note: defining MPFR_TESTS_DIVBYZERO without MPFR_ERRDIVZERO is just for debugging. (r11493) Vincent Lefèvre
08:03:19 Source Code scm commit: more work on MPFR_TESTS_EXCEPTIONS (r11492) Paul Zimmermann
07:29:24 Source Code scm commit: [tests/tests.c] Re-added code testing FE_DIVBYZERO and FE_INVALID with -DMPFR_TESTS_DIVBYZERO -DMPFR_ERRDIVZERO, because MPFR_TESTS_EXCEPTIONS is not ready yet. (r11491) Vincent Lefèvre
07:04:23 Source Code scm commit: [tests/tversion.c] Output IEEE floats status. (r11490) Vincent Lefèvre
06:09:23 Source Code scm commit: check for fp exceptions (experimental) (r11489) Paul Zimmermann
2017-05-17
14:08:40 Source Code scm commit: [src/cmp2.c] Added/updated comments. (r11488) Vincent Lefèvre
13:34:28 Source Code scm commit: [tests/tcmp2.c] Added tests with equal numbers, possibly of different sizes. (r11487) Vincent Lefèvre
10:15:30 Source Code scm commit: [src/cmp2.c] Corrected comments. (r11486) Vincent Lefèvre
06:46:27 Source Code scm commit: [tools/mpfrlint] In the detection of reserved macro names, skip ERR from src/mpfr-gmp.h (this is a special case). (r11485) Vincent Lefèvre
06:31:23 Source Code scm commit: [tests/tget_set_d64.c] Made PRINT_ERR_MISC more robust. (r11484) Vincent Lefèvre
06:27:14 Source Code scm commit: [tests/*.c] Renamed the remaining macros whose names are reserved by ISO C when <errno.h> is included. (r11483) Vincent Lefèvre
2017-05-16
13:44:56 Source Code scm commit: [tests/texceptions.c] Made code formatting consistent. (r11482) Vincent Lefèvre
13:41:49 Source Code scm commit: [tests/*.c] Renamed "ERROR" macro (macro name reserved by ISO C when <errno.h> is included) to "PRINT_ERROR", and made the macro more robust. (r11481) Vincent Lefèvre
13:21:15 Source Code scm commit: Define the EXP() macro in src/mpfr-gmp.h only when needed (macro name reserved by ISO C when <errno.h> is included). (r11480) Vincent Lefèvre
10:35:35 Source Code scm commit: [src/gamma.c] Got rid of the EXPM1_STR macro (macro name reserved by ISO C when <errno.h> is included), made no longer interesting by using mpfr_set_str_binary only once and another temporary variable. At the same time, replaced the usual allocation by a GROUP allocation (more efficient). (r11479) Vincent Lefèvre
2017-05-15
15:39:18 Source Code scm commit: [tests/tcheck.c] Code reformatting (GNU coding style). (r11478) Vincent Lefèvre
15:27:29 Source Code scm commit: [tests/tcheck.c] Renamed "ERROR" macro (macro name reserved by ISO C when <errno.h> is included) to "PRINT_ERROR". (r11477) Vincent Lefèvre
15:19:11 Source Code scm commit: Do not define macro names starting with "SIG" (+ uppercase letter) as they are reserved when <signal.h> is included (ISO C, 7.31.7), which happens to be the case in tests/tests.c, which also includes tests/mpfr-test.h and src/mpfr-impl.h indirectly. * src/mpfr-impl.h, tests/mpfr-test.h: renamed macros: SIGN → VSIGN SIGN_RAND → RAND_SIGN * {src,tests}/*.{c,h}: updated the names in the use of these macros. (r11476) Vincent Lefèvre
14:53:04 Source Code scm commit: [tools/mpfrlint] Added a comment about the definition of reserved identifiers (in particular, macro names). (r11475) Vincent Lefèvre
14:47:06 Source Code scm commit: [tools/mpfrlint] Detect the definition of reserved macro names. (r11474) Vincent Lefèvre
14:20:37 Source Code scm commit: [src/bernoulli.c] Changed isprime to is_prime in order to avoid obscure problems if <ctype.h> is added to mpfr-impl.h in the future (or if this function is made external). See ISO C "Future library directions", Clause 7.31.2 "Character handling <ctype.h>". (r11473) Vincent Lefèvre
2017-05-12
13:01:17 Source Code scm commit: [src/mpfr-impl.h] Improved comments about assertions and MPFR_ASSUME(). (r11472) Vincent Lefèvre
2017-05-09
15:33:20 Source Code scm commit: [INSTALL] Mention the problem when almost all the tests fails with a message "Incorrect MPFR version! ...". (r11471) Vincent Lefèvre
13:02:22 Source Code scm commit: Increase the buffer sizes even more, and added a comment. This may be better in case of quick future changes (possibly for debugging), without checking carefully. For the tests, optimal code is useless. (r11470) Vincent Lefèvre
10:47:36 Source Code scm commit: fixed bugs found by gcc 7.1.0 (-Werror=format-overflow=) [VL] The MPFR code was correct. This is actually a bug in GCC: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79257 (r11469) Paul Zimmermann
09:11:57 Source Code scm commit: [doc] Updated FAQ.html with update-faq. (r11468) Vincent Lefèvre
08:58:53 Source Code scm commit: [www/algo.html] Added my article "Correctly rounded arbitrary-precision floating-point summation". (r11467) Vincent Lefèvre
08:48:10 Source Code scm commit: [www/algo.html] Reorganization, with updated URL's. (r11466) Vincent Lefèvre
08:30:37 Source Code scm commit: [faq.xhtml,www/algo.html] TOMS article: use the stable doi.org URL; removed the obsolete link to the BibTeX entry. [www/faq.html] Regenerated with: xsltproc --nodtdattr faq-web.xsl faq.xhtml > www/faq.html (r11465) Vincent Lefèvre
08:22:14 Source Code scm commit: [www/pub.html] Updated a URL (https like the other ones). (r11464) Vincent Lefèvre
08:18:40 Source Code scm commit: [www/pub.html] Added a paper. (r11463) Vincent Lefèvre
2017-05-06
21:32:25 Source Code scm commit: [www/index.html] Added reference to Sirocco. (r11462) Vincent Lefèvre
2017-05-05
08:17:09 Source Code scm commit: [tools/mbench/timp.h] Updated TIMP_MEASURE() to handle the case where the return value is 0: this probably means that timp_overhead was too large and incorrect. In this case, TIMP_OVERHEAD() is called again to recompute timp_overhead and the timing is redone. Note: if we do not call TIMP_OVERHEAD() again, then TIMP_MEASURE() loops indefinitely, meaning that timp_overhead was really too large. (r11461) Vincent Lefèvre
07:42:06 Source Code scm commit: [tools/mbench/timp.h] Comment at the wrong place. Fixed. (r11460) Vincent Lefèvre
07:40:54 Source Code scm commit: [tools/mbench/timp.h] Added a comment about r11458. (r11459) Vincent Lefèvre
2017-05-04
17:07:27 Source Code scm commit: [tools/mbench/timp.h] Fixed overflows on unsigned arithmetic. (r11458) Vincent Lefèvre
16:56:59 Source Code scm commit: [tools/mbench/mfv5.h] Increased values from 2^60-1 to 2^64-1 (it is possible to get more than 2^60-1 cycles in practice!). (r11457) Vincent Lefèvre
09:40:05 Source Code scm commit: Merged the latest changes from the trunk, including some old changesets related to mpfr_zeta that were skipped, resolving conflicts. Added RNDF support to new code introduced by this merge: * mpfr_mul_1n in src/mul.c (from r11281); * mpfr_sqr_1n in src/sqr.c (from r11283); * mpfr_div_1n in src/div.c (from r11284); * mpfr_sqrt1n in src/sqrt.c (from r11293). (r11456) Vincent Lefèvre
08:38:17 Source Code scm commit: Merged r11198-11280 from the trunk; no conflicts but additional changes: * About r11271 (src/add1sp.c), which introduces new special code (function mpfr_add1sp1n), handle MPFR_RNDF in the same way as done in similar existing special code (mpfr_add1sp1 and mpfr_add1sp2). * In mpfr_add1sp3, do the same thing (this should have been done in r11172, where this function was introduced via a merge). * About r11279 (src/sub1sp.c, tests/tsub1sp.c), which introduces new special code (function mpfr_sub1sp1n), do the same thing. In tests/tsub1sp.c, s/RND_LOOP/RND_LOOP_NO_RNDF/ as usual to avoid a failure. * Note: concerning mpfr_sub1sp3, RNDF support was added at the same time of the merge in r11179. * Some style changes related to RNDF, in particular for consistency. (r11455) Vincent Lefèvre
07:55:17 Source Code scm commit: Merged r11197 from the trunk, resolving the conflict. (r11454) Vincent Lefèvre
07:46:57 Source Code scm commit: Merged r11179-11196 from the trunk (no conflicts). (r11453) Vincent Lefèvre
2017-05-02
14:28:40 Source Code scm commit: [src/gamma.c] Clarified proof for tiny arguments again. (r11452) Vincent Lefèvre
14:07:25 Source Code scm commit: [src/gamma.c] Clarified proof for tiny arguments. (r11451) Vincent Lefèvre
12:38:02 Source Code scm commit: [src/gamma.c] In the proof for tiny arguments: * Simplification suggested by Paul. * Attempt to clarify the end. (r11450) Vincent Lefèvre
09:23:25 Source Code scm commit: added comment (r11449) Paul Zimmermann
08:10:53 Source Code scm commit: added reference for mbench (rdtsc vs rdtscp) (r11448) Paul Zimmermann
2017-04-28
15:41:36 Source Code scm commit: [src/gamma.c] Tiny arguments: corrected a typo and added a TODO (one might get a better condition). (r11447) Vincent Lefèvre
15:19:16 Source Code scm commit: [src/gamma.c] Corrections in a proof for tiny arguments. Added a TODO. (r11446) Vincent Lefèvre