diff options
Diffstat (limited to 'ChangeLog')
| -rw-r--r-- | ChangeLog | 168 |
1 files changed, 167 insertions, 1 deletions
@@ -82,13 +82,179 @@ * eval.c (set_LINT): Reset lintfunc to `warning' for LINT="invalid". Thanks to Andy Schorr for the report. +2016-07-08 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h: Restore previous comment about unterminated strings, since + I am removing the string termination patches from field.c + (free_api_string_copies): Declare new gawkapi function. + * builtin.c (do_mktime, do_system): Restore temporary string + termination to protect against unterminated field values. + (nondec2awknum): Remove comment about unnecessary termination. + * eval.c (posix_compare): Restore temporary string termination. + * field.c (databuf): Remove struct no longer needed. + (set_field): Remove memcpy for string termination, since we will support + unterminated field string values. + (rebuild_record): Ditto. Also no need to allocate space for terminated + string copies. + (allocate_databuf): Remove function, since memory management can again + be done inside set_record. + (set_record): Restore inline buffer management logic. + (reset_record): Remove calls to allocate_databuf, since we no longer + need space for making terminated copies of field strings. + * gawkapi.c (free_api_string_copies): New function to free strings + that we made to provide terminated copies to API functions. + (assign_string): New function to convert a string to an awk_value, + making sure to copy it if we need to terminate it. + (node_to_awk_value): Use assign_string to return string values with + NUL termination protection. + * int_array.c (is_integer): Restore temporary string termination. + * interpret.h (Op_push_i): Ditto. + (Op_ext_builtin): After external function returns, call + free_api_string_copies to free temporary string copies. + * mpfr.c (force_mpnum): Restore temporary string termination. + * node.c (r_force_number, get_ieee_magic_val): Ditto. + 2016-07-08 Arnold D. Robbins <arnold@skeeve.com> * dfa.c: Sync with GNU grep. Unrelated: - * builtin.c (do_print): Coding style change. + * builtin.c (do_print): Coding style change. (This change obsoleted + by earlier changes in the fixtype branch.) + +2016-07-06 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h: Modify comments to indicate that MAYBE_NUM will now be + left enabled to indicate strnum values by the NUMBER|MAYBE_NUM + combination, whereas STRING|MAYBE_NUM indicates a potential strnum. + (fixtype): Modify MAYBE_NUM test to avoid calling force_number if + NUMCUR is already set. + * builtin.c (do_typeof): Call fixtype to resolve argument type. + This forces parsing of numeric strings, so there's a performance + penalty, but we must do this to give a correct result. The meaning + of "strnum" changes from "potential strnum" to "actual strnum". + * eval.c (set_TEXTDOMAIN): Remove some dead code left over from last + patch. + * int_array.c (is_integer): When a MAYBE_NUM is converted successfully + to a NUMBER, leave the MAYBE_NUM flag enabled. + * mpfr.c (mpg_force_number): Ditto. + * node.c (r_force_number): Ditto. + +2016-07-06 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h: Modify stptr comment to indicate that all strings are now + NUL-terminated. + * builtin.c (do_mktime): Remove unnecessary logic to terminate + the string with '\0' temporarily. + (do_system) Ditto. + (nondec2awknum): Add a comment about termination. + * eval.c (posix_compare): Remove logic to terminate strings temporarily. + (set_ORS): No need to terminate ORS, since the string node is already + terminated. What gave us the right to modify that node anyway? + (fmt_index): Remove code to terminate string. This seems to have been + invalid anyway, since we don't own that memory. + (set_TEXTDOMAIN): Do not terminate TEXTDOMAIN string, since the node + is already terminated. We didn't have the right to modify that node + anyway. + * gawkapi.c (node_to_awk_value): Add assert checks to confirm that the + string is NUL-terminated. + * gawkapi.h: Modify awk_string comment to indicate that strings are + always terminated with '\0'. + * int_array.c (isinteger): Remove unnecessary logic to terminate string + with '\0' temporarily. + * interpret.h (Op_push_i): Ditto. + * io.c (nextfile): Remove string termination. We didn't own that memory + anyway. + * mpfr.c (force_mpnum): Remove unnecessary logic to terminate the + string with '\0' temporarily. + * node.c (r_force_number): Remove NUL termination around strtod call, + since we already know that there is either a white space or '\0' + character there. Either one will stop strtod. + (get_ieee_magic_val): Ditto. + * profile.c (pp_number): No need to terminate string returned by + r_format_val. + +2016-07-06 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * interpret.h (Op_field_spec): Now that all $n field values are + NUL-terminated, there is no reason to call dupnode for $n where n > 0. + This saves malloc and copying overhead, thereby more than offsetting the + performance hit of the additional copying and NUL-termination in the + last patch to field.c. It also eliminates repeated parsing in cases + where $n, for n > 1, was accessed more than once in a numeric context, + so the new approach should be a performance win. + +2016-07-06 Andrew J. Schorr <aschorr@telemetry-investments.com> + + Make sure that all field values, and therefore all strings inside gawk, + are terminated with a '\0' character! + * field.c (databuf): New static struct to hold info about our buffer to + contain the field string values. + (allocate_databuf): New function to make sure the databuf is large + enough to hold $0 and copies of $1 through $NF. + (set_field): Copy $n into free space previously allocated in databuf + and add a '\0' at the end. + (rebuild_record): Call allocate_databuf to ensure sufficient space + for copying non-malloced field values. When copying field values, + use databuf to create a NUL-terminated copy. + (purge_record): New function extracted from reset_record to initialize + $1 through $NF to null values. + (set_record): Buffer management moved to new allocate_databuf function. + Call purge_record instead of reset_record, since reset_record contains + some extra logic not needed in this case. + (reset_record): Call purge_record to do most of the work, and call + allocate_databuf to make sure we have a big enough buffer to contain + copies of the $1 through $NF. + +2016-07-06 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h: Renumber flags to remove gap created when FIELD was removed. + +2016-07-05 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * field.c (rebuild_record): Need to set MALLOC flag if we allocate + memory for a residual field node with valref > 1. + +2016-07-05 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * field.c (rebuild_record): Do not bother to create new field nodes + to replace malloc'ed nodes when rebuilding $0. + +2016-07-05 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h (FIELD): Remove unnecessary flag. + (MALLOC): Move definition to join the others, and improve the comment. + * array.c (value_info): Replace FIELD test with MALLOC test. + * eval.c (flags2str): Remove FIELD flag. + * field.c (init_fields): Remove FIELD bit from Null_field->flags. + (set_field): Remove FIELD bit from flags. + (rebuild_record): Test against MALLOC instead of FIELD. If a field + node has valref > 1, we should make a copy, although I don't think + it is valid for this to happen. + (set_record): Remove FIELD bit from flags. + * interpret.h (UNFIELD): Add comment, and test MALLOC flag instead of + FIELD. Remove probably buggy code to disable the FIELD flag when + valref is 1; that would have created a node where neither the FIELD + nor MALLOC flag was set, which seems invalid. + * node.c (r_dupnode): Remove code disabling FIELD flag. + +2016-07-04 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * awk.h (force_string_fmt): New inline function to get the string + representation in a requested format. + (force_string): Reimplement as a macro using force_string_fmt function. + (force_string_ofmt): New macro to get a value's OFMT representation. + * builtin.c (do_print): Use new force_string_ofmt macro instead of + duplicating the logic inline. + +2016-07-04 Andrew J. Schorr <aschorr@telemetry-investments.com> + + * str_array.c (str_lookup): There is no need to worry about the + MAYBE_NUM flag, since the code has been patched to make sure to + preserve the string value of strnum values, and the integer array + code should no longer mistakenly claim a strnum integer with a + nonstandard string representation. 2016-07-03 Andrew J. Schorr <aschorr@telemetry-investments.com> |