open opcode returns true instead of returning an exception

[dukeleto]

This test should not pass:

.include 'test_more.pir'
.local pmc pipe
pipe = open 'idontexist', 'rp'
ok(pipe,'pipe is true')

I ran into this testing Plumage from NQP. I have only verified that this occurs on darwin/x86. I will check other platforms.

[dukeleto]

I used the Parrot Shell to verify this. In the root of a parrot repo, type : perl tools/dev/parrot_shell.pl then paste the code snippet :

parrot_shell 4> .include 'test_more.pir'
.local pmc pipe
pipe = open 'idontexist', 'rp'
ok(pipe,'pipe is true')
.
Time: 0.0564442 wallclock secs 
Output:
ok 1 - pipe is true

[NotFound]

An exception in open is only expected for exceptional conditions, like being unable to create a handle, and in the pipe's case being unable to fork or to launch a process, for example.

The pipe ask the operating system to do some work. That work can fail at some later point. get_bool on FileHandle objects don't check error conditions, just check if the object already is in error state.

Current pipe implementation just launch a shell, the shell can start without problem even if the program you ask it to execute does not exist, it takes some time to search the PATH and such. Waiting for stablization or something in open or in get_bool will compromise the possible implementations, probalbly slowing them down.

So, I don't think this is a bug, just wrong expectations.

[dukeleto]

Does a test like this exist already? If not, where should this one live?

[NotFound]

A test like what? You want to test that trying to execute a non existent program fails? Well, fails at doing what? It can hardly fails to do nothing.

If you want to check the exit value of the launched process or something like that, that is highly platform dependent.

[dukeleto]

Thank you for the clarification. This can be chalked up to a bug in the way Plumage uses Parrot pipes, so this ticket can be closed.