The fix was simple, but hard to find. I had to use CommonMark with
extensions because apparently, Pandoc does not include definition lists
in the normal CommonMark.
Commit e9c191cdda supposedly adds support
for `SIGSTOP` in history. However, what it does is add support for
`SIGTSTP` and then uses `SIGSTOP` to simulate it. This commit uses the
correct terminal codes for each and raises the correct signal for each.
This bug was reported in #10. It
happens when the first thing a user enters at `stdin` causes a parse
error.
mathieu did not know what caused the error, and I could not reproduce it
on my machine. I tried on FreeBSD and could reproduce it under root, but
not my user.
After investigating, the reason I could not reproduce it was because my
user runs a file: /home/gavin/.bcrc. Because it runs the file, it would
execute the `scale = 20` statement in the math library. However, when I
would not have it execute that file, and enter a parse error first, that
statement would not be executed, and since bc resets after an error,
that statement would disappear and never happen.
The fix was simple: just call bc_program_exec() right after loading the
math libraries.
Fix#10
When I first added the test, I tried to use cut(1) because it is in
POSIX. However, I couldn't figure out how it worked. (The POSIX standard
was not well written for it.) That was why I used head(1). This patch
changes the test to use cut(1) instead.
This was contributed by the OpenBSD user and contributor Brian Callahan.
Thank you!
I want to change this because I was using a print statement to track
progress in a script, and it wasn't flushing. It only flushed when I
sent a SIGINT.
That is not good behavior, and it's behavior I have only because I had
to implement my own flushing for bc.
Fix this by flushing when stuff is printed.
This requires some code to make things match POSIX and especially making
inline an empty string because Microsoft's compiler is picky about
functions marked inline.
This bug was hard to find the cause of (because the reproducer was a
massive dc script), but the problem was simple: the `P` was supposed to
pop the top of the stack, and it did not.
This commit fixes the problem.
The problem was not that stuff was printed without newlines, it was that
history did not recognize it as such. This commit fixes that.
To do that, I had a way for bc_file_* functions to know what kind of
flush they needed to do. I added an enum for it and updated all of the
places. This ensures that there isn't an infinite loop or other such
problem when flushing during history.
This commit does not complete the change. I have to ensure it builds
with history turned off still.
Gavin Howard
Brian Callahan