"Overwrite Old Files", and revamp that section to make it easier to
follow. "tar" -> "GNU tar" where appropriate. Migrate getdate
documentation into getdate.texi. Fix several minor typos. Describe
TAR_OPTIONS. Describe incompatibility between incremental backups and
--atime-preserve. Describe incompatibility between --verify and other
options. Mention that tar normally removes symbolic links rather than
following them, when extracting a file of the same name.
(struct delayed_symlink): New type.
(delayed_symlink_head): New var.
(extr_init, fatal_exit):
Invoke extract_finish on fatal errors, not apply_delayed_set_stat.
(set_mode, set_stat): Pointer args are now const pointers.
(check_time): New function.
(set_stat): Warn if setting a file's timestamp to be the future.
(make_directories): Do not save and restore errno.
(maybe_recoverable): Set errno to ENOENT if we cannot make missing
intermediate directories.
(extract_archive):
Invoke apply_nonancestor_delayed_set_stat here, not in caller.
Extract potentially dangerous symbolic links more carefully, deferring their
creation until the end, and using a regular file placeholder in the meantime.
Do not remove trailing / and /. from file names.
Do not bother checking for ".." when checking whether a directory
loops back on itself, as loopbacks can occur with symlinks too.
Also, in that case, do not bother saving and restoring errno; just
set it to EEXIST.
(apply_nonancestor_delayed_set_stat): A prefix is a potential ancestor if
it ends in slash too (as well as ending in a char just before slash).
(apply_delayed_set_stat): Remove.
(apply_delayed_symlinks, extract_finish): New functions.
Do not invoke apply_nonancestor_delayed_set_stat;
DO_SOMETHING is now responsible for that.
Do not invoke apply_delayed_set_stat; our caller is now responsible for that.
(read_header): Use signed char instead of signed_char.
Prevent later references to current_header from
mistakenly treating it as an old GNU header.
(from_header): Quote invalid base-64 strings in diagnostics.
(time_from_header): Do not warn about future timestamps in archive;
check_time now does that.
(tartime): Fix off-by-one error when copying year if OLD_CTIME.
(print_header): Quote unknown file types.
(skip_member): New function, replacing skip_extended_headers and
now skipping the whole member instead of just the extended headers.
All callers changed. This makes the code handle extended headers
uniformly, and fixes some bugs.
(child_open_for_compress, child_open_for_uncompress, close_archive):
Propagate any failure of the compression process back to "tar".
(open_archive, flush_write, flush_read, close_archive):
Do not allocate an array of size PATH_MAX, as PATH_MAX might be (size_t) -1.
Instead, allocate an array with the size that's needed.
(open_archive): Don't bother checking S_ISCHR of /dev/null.
(backspace_output): Don't try to backspace past start of archive.
(close_archive): Remove special case for DELETE_SUBCOMMAND.
Include <hash.h>.
(getpwuid, getgrgid): Declare only if system headers don't.
(gid_to_gname): Don't invoke setgrent.
(namelist): Now static, not global.
(nametail): New var. All uses of namelast changed to use nametail,
with one extra level of indirection.
(name_gather): Use memcpy instead of strncpy + assignment of NUL.
(name_match): Set nametail too, when setting namelist to null.
(add_hierarchy_to_namelist): Change type of dir arg from char * to
struct name *, so that we don't have to look up the name again here.
Get change_dir from dir rather than as a separate arg. Add dirsize arg,
and pass it along to get_directory_contents. Remove unnecessary check
of directory type.
(new_name): Do not append a slash if PATH already ends in one.
(avoided_names, struct avoided_name): Remove.
(avoided_name_table): New var, replacing avoided_names.
(hash_avoided_name, compare_avoided_names): New function.
(add_avoided_name, is_avoided_name): Use hash table rather than linked list.
(NAMLEN): Remove macro.
(malloc, realloc): Remove decls.
(stpcpy): Likewise.
("xalloc.h"): Include.
(NAME_SIZE_DEFAULT): New macro.
(savedir): Use xmalloc / xrealloc to allocate memory.
Use NAME_SIZE_DEFAULT if name_size is negative or overflows to zero.
Skip "" directory entries.
Use strlen to calculate directory entry length, since the old method
is rarely used these days and isn't worth supporting.
Check for integer overflow when calculating allocation size.
Use memcpy to copy entries, instead of stpcpy.
Set errno properly when returning NULL.
Check for readdir error.
(textint): New typedef.
(parser_control): Changed from struct parser_control to typedef (for
consistency). Member year changed from int to textint. All uses changed.
(YYSTYPE): Removed; replaced by %union with int and textint members.
(tID): Removed; not used.
(tDAY, tDAY_UNIT, tDAYZONE, tHOUR_UNIT, tID, tLOCAL_ZONE, tMERIDIAN,
tMINUTE_UNIT, tMONTH, tMONTH_UNIT tSEC_UNIT, tSNUMBER, tUNUMBER,
tYEAR_UNIT, tZONE, o_merid): Now of type <intval>.
(tSNUMBER, tUNUMBER): Now of type <textintval>.
(date, number, to_year): Use width of number in digits, not its value,
to determine whether it's a 2-digit year, or a 2-digit time.
(yylex): Store number of digits of numeric tokens.
Return '?' for unknown identifiers, rather than (unused) tID.
after handling them.
(list_archive): Use consistent message for unexpected EOF.
Use error message functions to report errors consistently.
(read_header): Use xalloc_die to report memory exhaustion.
Use consistent message for unexpected EOF.
(decode_mode): Remove; moved to misc.c.
(skip_file): Use consistent message for unexpected EOF.
(skip_extended_headers): Likewise.
All callers changed.
(deal_with_sparse): Don't keep reading after read errors.
(finish_sparse_file): Just abort if there is an internal error.
Use error message functions to report errors consistently.
(dump_file): Fix typo: stat_warn and stat_error were interchanged.
Quote file names with colons if possible.
Don't restore access times on directories during incremental dumps
until after dealing with the directory.
If ignoring failed reads, count closedir errors as warnings, not errors.
Fix buffer overrun problem when dumping sparse files.
If ignoring failed reads, count read errors as warnings, not errors.
Use error message functions to report errors consistently.
If ignoring failed reads, count unknown files as warnings, not errors.
