add tests for all bash prompt special characters

[simonLeary42]

#2235

[simonLeary42]

is it ok to rely on external programs basename and tty?

[andychu]

Thanks, this looks very helpful!

I let the CI run, but the spec tests timed out after 15 minutes. Then I cloned locally and noticed that test case 27 for \j hangs on my machine ?

If I do

test/spec.sh prompt -r 27 

then it hangs

---

Though I am not sure why, because when I run it manually it doesn't hang

$ cat _tmp/27
#### \j for number of jobs
PS1='foo \j bar'
echo "${PS1@P}" | egrep -q 'foo 0 bar'
echo matched=$?
sleep 999 &
echo "${PS1@P}" | egrep -q 'foo 1 bar'
echo matched=$?
kill %%
fg
echo "${PS1@P}" | egrep -q 'foo 0 bar'



andy@hoover:~/git/oils-for-unix/simonLeary42/oils$ bash _tmp/27
matched=0
matched=0
_tmp/27: line 9: fg: no job control

[andychu]

Here is the CI dashboard

https://mb.oilshell.org/uuu/github-jobs/9010/

and then if you click through you can see a timeout after 15 minutes, which is the max we let it run:

https://mb.oilshell.org/uuu/github-jobs/9010/#job-ovm-tarball

---

The tests aren't really sandboxed, so we basically use a Docker container I build as a source of truth ... (could be improved at some point)

---

To make progress, I might just comment out test case 27 for now, and then we can see if the others pass?

If there are a bunch of green tests for bash, and a red ones for OSH, and it runs reliably in the CI, then that is good progress, and can be merged

---

I tend to break things up in steps, so we can do \j last I guess

https://github.com/oils-for-unix/oils/wiki/Spec-Tests

[andychu]

BTW the way I "extract" the test case is test/spec.sh prompt -r 27 -p > _tmp/27 -- that prints it instead of running it

And I either run with bash on my machine, or a specific versions of bash that we have in build/dev-shell.sh , i.e. compiled from source

---

How many of the tests pass locally on bash, on your machine? I am seeing a few failures other than the \j one, which I haven't diagnosed yet

[simonLeary42]

$ sleep 999 &
[%1] PID 12220 Started
$ echo %%
%%
$ kill %%
kill: ‘%%’: invalid process id
$ fg
fg: PID 12220 Continued
^C
$

since %% isn't expanded since osh doesn't provide a kill shell builtin that supports job IDs with the % syntax, osh pulls the sleep 999 process into foreground and sleeps for 999 seconds. I just turned down the sleep time to 5 seconds

[simonLeary42]

I'm struggling to come up with a good way to run the "command number" prompt test. I have to either find a reliable way to get the current "command number", which I'm not sure is even implemented in osh, and then just test that the next "command number" is the previous "command number" +1, or find a reliable way to spawn a subprocess of the current shell. I can't use $0 because -bash isn't a valid path, I can't use $SHELL because osh doesn't set that variable as far as I can tell, and the "command number" doesn't reset when using parentheses. I could try and use other clues to figure out what the current shell is and then hard code the path to each shell, but I don't like that...

[simonLeary42]

on my machine, all the prompt test cases pass for bash except 0 and 32.

[andychu]

OK looks like 9 tests fail on OSH, and they pass on bash:

https://mb.oilshell.org/uuu/github-jobs/9023/ovm-tarball.wwz/_tmp/spec/osh-py/prompt.html

Thanks!

I pushed a change to add this metadata line, which lets the CI pass

## oils_failures_allowed: 9

and then we can "whittle down" the errors as we go

[andychu]

I also added you as a collaborator, so you can run the CI yourself, and don't have to wait for me to hit "approve"

Most contributors directly modify a branch of oils-for-unix/oils for that reason, instead of having their own fork

It is a weird github limitation - https://github.com/oils-for-unix/oils/wiki/Contributing

[andychu]

OK let me see what hapepns if I try to merge ... the CI checks didn't finish because of the permission issue, but maybe it will go through on soil-staging (confused about github's rules)

[andychu]

(The merged worked! But from now on, please work in the upstream repo, so then you can see the CI results immediately without waiting for approval )

---

Thanks for the tests! I was able to knock off \v and \V very quickly

b2da0db

And it worked in C++ transparently

Let me look at some others

[andychu]

I think \D{} is actually more general than \t \T \@ \d

I mean I can understand why we should have them, but it annoys me to carry that over, since it is more readable to use \D{} and an explicit format IMO

rather than remembering what those mean

[andychu]

I am not quite sure how to get history number or command number, and what the exact difference is

I think the history number should come from GNU readline -- if we already have an API, then it shouldn't be too hard to add