gavinhoward
commented
1 year ago i haven't been able to update for some time (5.0.1 was our last update!)
Oof...I'm sorry. Have I been missing communication with you?
because the tests fail in a weird way without much useful information. i've included the current failures for 6.2.2 below, for example.
I agree; that is useless.
my guess is that this is something to do with Android using mksh as its /bin/sh. the interleaved output makes it appear like there might be some disagreement about job handling between bash and mksh?
I think that's a great guess, and it's my guess too because the tests/all.sh script does use job control. I tried to write it according to the POSIX standard, but I suspect that job control is finicky in general.
are the current bc tests known to work on shells other than bash (and in particular mksh)? any ideas where i should look (or some kind of parallelism i can just disable)?
They are known to work on tcsh (FreeBSD), but other than that, bash, and dash, I don't know.
any ideas where i should look (or some kind of parallelism i can just disable)?
To disable parallelism, change the last line of run-bc-tests-on-android.sh to:
exec adb shell $dash_t /data/local/tmp/bc-tests/tests/all.sh -n bc 0 1 0 0 0 bcI made two changes:
- Most importantly, there is a
-noption, which is how to disable parallelism in thetests/all.shscript. - I also added another 0 before the last
bcbecause there are now five integer positional arguments.
My hope (and it is only a hope) is that that missing integer argument is what is causing the weird failures because the script would otherwise treat the last bc argument as the last integer argument and cause some sort of failure. bash fails because it has a non-integer argument, but I don't know how mksh would fail in that case.
The other reason there might be failures is because I added a test that is "problematic." It's my test for malloc() failure, and unfortunately, it only works on Linux systems with swap disabled and some other changes. I don't think it would work on Android. I did add an option to disable that test in tests/all.sh; that is the missing integer option, in fact. But I don't think that's the reason because I added it in the fourth position (the last is whether to time the tests), so it is a 0 in your current script.
Regardless, with the parallelism disabled, you should see exactly what test is failing, and you should be able to give me that output, even if either of those are not the reason for the failure.
In the meantime, I'll work on my end. Android is a first-class user of my bc, and I need to start treating it that way, including setting up a building and testing environment. I'll do my best to do so myself, but if possible, can you give me a helpful link?
enh-google
i haven't been able to update for some time (5.0.1 was our last update!) because the tests fail in a weird way without much useful information. i've included the current failures for 6.2.2 below, for example.
my guess is that this is something to do with Android using mksh as its /bin/sh. the interleaved output makes it appear like there might be some disagreement about job handling between bash and mksh?
are the current bc tests known to work on shells other than bash (and in particular mksh)? any ideas where i should look (or some kind of parallelism i can just disable)?