POSIX mandates that if the input of rm is a tty and it does not have
write rights over a file/dir then it should ask for confirmation, in
the same way that is done with the -i flag. To accomodate both things
the code has been rearrenged a bit to have only one case instead of
having two. Also, this rework adds the error message when a directory
is removed without a -r flag.
unescape() is used by several tools, in particular printf(1) and
tr(1), which should stop the octal escape at a maximum of 3 digits:
printf(1)
> In addition to the escape sequences shown in XBD 5. File Format
> Notation ('\\', '\a', '\b', '\f', '\n', '\r', '\t', '\v'), "\ddd",
> where ddd is a one, two, or three-digit octal number, shall be
> written as a byte with the numeric value specified by the octal
> number.
tr(1)
> An octal sequence shall consist of a <backslash> followed by the
> longest sequence of one, two, or three-octal-digit characters.
Previously, the maximum was set to 4 (possibly a typo?), which meant
that printf '\0123' printed `S` instead of `<newline>3`.
To check that this doesn't break any other tools using unescape:
- cut: used for -d parameter, escapes are non-standard
- join: used for -t parameter, escapes are non-standard
- nl: used for -s parameter, escapes are non-standard
- paste: used for -d parameter, POSIX specifies \n, \t, \\, and \0,
\0 followed by a digit is unspecified
- sort: used for -t parameter, escapes are non-standard
Quoting POSIX:
The following exit values shall be returned: 0 All requested files (excluding files where a non-affirmative response was given to a request for confirmation) were successfully copied. >0 An error occurred.
As requested, I resend my old patch for fixing the crashing while
archiving with names longer than 100 characters.
Last patch dealing with the issue was [1], and the old patch was [2]. The
code before this commit was not dealing correctly with multiple slashes,
but use of basename(3) and dirname(3) needed a temporary buffer because
otherwise we destroyed the path that was used later in several places.
This solution does not modifies the path and use pointer arithmetic to
solve the problem.
[1] https://lists.suckless.org/hackers/2412/19213.html
[2] https://lists.suckless.org/hackers/2402/19071.html
Co-authored-by: Roberto E. Vargas Caballer <k0ga@shike2.net>
The Makefile was relaying in two extensions:
1- Inference rule chaining: Based in getconf.c -> getconf.o -> getconf
2- Inference rule .o: POSIX defines the .c rule, but not the .o rule
(what makes sense because to be useful you need inference rule chaining).
We are using a wildcard to copy things into the directory used to
build the distribution tar file, and this directory can contain
the .1 extension used in the *.1 wildcard used to copy the man
pages. Also, changed the cp command line from -r to -R because
-r is not specified in POSIX.
TRAP is a useful signal which when sent to a process currently
attached to a debugger causes the debugger to immediately break.
This is very helpful for programs running in some sort of run loop
where it may be difficult to set up a break point in an
appropriate location without modifying the source to include a
predicate condition.
It would be a good idea to add a usage instruction for sbase-box
in sbase's README similar to the one in ubase's README.
Also, There's no command named 'sha238sum' but there is one named
'sha384sum' and in the alphabetized list of commands 'paste' should
come before 'pathchk'.
When viewing sbase's man pages the date displayed at the bottom is
the current date rather than the date entered in the <program>.1
file.
According to this:
https://mandoc.bsd.lv/mdoc/details/date.html
The traditional man date format used by sbase (.Dd year-month-day)
is no longer recommended and no longer supported by GNU troff or
Heirloom Docs and as a result the current date is displayed instead.
Although this format is still accepted by mandoc for backward
compatibility.
For portability it states the standard mdoc date format should be
used instead: .Dd month day, year
I applied the following patch to sbase's last commit b30fb56 and
it fixes the issue.
The target tried to execute a non valid shell script
and there was a possibility to fatorize the creation
of the proto file instead of repeating it in all
the actions that required.
Some tar archives (eg. ftp://ftp.gnu.org/gnu/shtool/shtool-2.0.8.tar.gz)
use leading spaces instead of leading zeroes for numeric fields.
Although it is not allowed by the ustar specification, most tar
implementations recognize it as correct. But since 3ef6d4e4, we
replace all spaces by NULs here, not just trailing ones, which leads to
recognizing such archives as malformed. This fixes it: we now skip
over leading spaces, allowing strtol(3) to read those numeric fields.
As pointed out in a mail to dev expr was segfaulting when multiple
math operations were specified on the command line: eg. 'expr 3 \*
2 + 1'. This happens because the tonum(), introduced in e50d533,
assumed that v->str was always non null. parse() guarantees this
for user input but this is not the case when doop() is called with
the result of a previous calculation. However in that case we know
that v->num is already valid so we can simply return.
The MAN macro was not used, and it and OBJ had the same problem
because they were defined using an empty string in the replace
pattern of the macro expansion, but as it is said by POSIX:
The subst1 to be replaced shall be recognized when it is a suffix
at the end of a word in string1
so, an empty string should not be used.
Also, a new inference rule is added to generate the binary
directly from the .c without generating the intermediate
object, removing the need of chaining different inference
rules which is not guaranteed to work in all the make
implementations.
POSIX specifies that if the pattern contains a subexpression then
the first matched subexpression should be returned if it exists.
This fixes things like the following:
./expr 00003 : '\(.*\)'
Before: 3
After: 00003
Comparison operations (>, <, =, etc.) and matching operations must
operate originally provided string not one that has gone back and
forth through string formatting. This caused operations such as
the following to give incorrect results:
./expr 00003 : '.*'
Before: 1
After: 5
This commit fixes that issue.
The s command can apply a replace pattern with embedded newlines
which modifies the line/index assignament. Using a range in the
address fail because afther the call to subline() the next line
has to be searched based in the index because the replace could
insert newlines.
when sbase-box is executed without argument, the check sbase-box
options doesn't verify the argument count leading to a segfault.
Add a check on the argc before parsing sbase-box options (currently
only `-i`)
If newcmd is 0 then error() undo all the modifications
that happened since the last command, but this is not
what POSIX mandates:
SIGINT The ed utility shall interrupt its current activity, write the
string "?\n" to standard output, and return to command mode
(see the EXTENDED DESCRIPTION section).
Strings without newlines created problems in the function
and the global field was not updated, making that new lines
added were marked as global being processed in the current
global command.
If the string r.str is freed but error() is called then
next call will see a pointer that maybe it will try to free
because the call to error unwind the frame stack.
These functions allow to read from stdin the full next
line or seting as input a character array. These functions
avoid all the complexity about repeat commands that is very
fragile and depends on having multiple global variables with
weak relation between them.
