The signal handlers were calling longjmp() but as the code was calling
non signal safe functions the behaviour was very unpredictable generating
segmentation faults and dead lock. This commit changes the signal handlers
to only set a variable that is checked in safe places where long loops
happen.
POSIX.1-2017 demands in Shell & Utilities under 'Commands in ed':
The e, E, f, r, and w commands shall take an optional file parameter,
separated from the command letter by one or more <blank> characters.
Ensure at least one <blank> character (as defined for the POSIX locale)
is present or error out.
Signed-off-by: Rene Kita <mail@rkta.de>
Fopen() and Popen() were open as read streams, but we were writing
in both cases. In the same way, the FILE pointer returned by popen()
was close with fclose() that can lead to file descriptor leaks and
zombie processes.
Discard() was reading stdin until a new line was found, but in
case of having an empty line in the input buffer then it didn't
make sense because we were just discarding the full next line.
When the file name begins with ! then the addressed buffer content is
piped to the name of the file that is considered a command that is
executed in a shell.
After join() is called for the first time, s.str is left pointing
to a string that was just freed. Upon the second call to join(),
it is freed again at the start of the function.
Since the string is reset on every function call, there is no reason
for it to be static, so just replace the initial free with assignment
to NULL.
This expression was wrong, but it was causing a false positive
in some compilers that couldn't see that error() cannot return.
The actual problem of the line is that it was too complex and it is better
to split it in simplex expressions.
Copy was using directly the line numbers and incrementing them
without calling nextln(). It also didn't worry about how
line numbers are modified when we insert new lines.
Current handling of strings is a bit messy. This type is copied
from the sed implementation. Addchar_ is added to be able to live
with String and old style chars based in 3 different variables.
From 6665eaa1d2c25a95b44a4f4fb3d24a3bd5c1180f Mon Sep 17 00:00:00 2001
From: Thomas Mannay <audiobarrier@openmailbox.org>
Date: Thu, 3 Nov 2016 15:16:32 +0000
Subject: [PATCH] Treat addresses of 0 as 1 for insert
By stripping backslashes this code caused a number of bugs.
'\<digit>' expressions caused literal <digit>s to be subbed-in,
'\&' was treated identically to '&', and other escaped characters
added garbage to the string.
regexec(3) happily matches /^$/ against the text of line
zero (the null string), causing an error.
Also update email address for Wolfgang Corcoran-Mathe
Given:
static int marks['z' - 'a'];
marks[c];
In this case c is the character read and index is out of bounds. We
need marks[c - 'a']
While playing around with optimization settings gcc caught caught this.
Also fix one check for c, change from isalpha to islower.
-emg
From 2cc36818283e9068576c1042690c016a81b709a3 Mon Sep 17 00:00:00 2001
From: Evan Gates <evan.gates@gmail.com>
Date: Fri, 15 Jul 2016 09:52:39 -0700
Subject: [PATCH] fix marks indexing
It is impossible to rematch a pattern which has one (or both)
of these operators, so the simplest solucion is detect them
while we are compiling the regular expression and break the
match loop after the first iteration.
Ed was falling doing substitution different of the first or all
(s//%/, s//%/\1, s//%/g), because it was not adding the matches
which were not going to be substituted.
Rematch() was incremnenting the last match always, even in the
cases when it was not matching anything. This patch detects
this situation and it updates it only when there is a match.
