> On 3 Jan 2017, at 06:58, Joachim Durchholz <j...@durchholz.org> wrote: > Yes, chdir is nasty because it's a global setting and you never know which of > the functions you're calling might be changing the setting. > > indir solves all these things. (Provided it changes the current directory > back even in the face of exceptions.)
Please note that there is no real chdir() at the OS level, because that is a bad idea in a threaded environment. And for that reason, the JVM doesn’t allow for it, afaik, and thus it wouldn’t work on the JVM backend anyway. The “current directory” is represented by the $*CWD dynamic variable. Changing the “current directory” in a scope, is nothing else than syntactic sugar around creating a local lexical $*CWD. By virtue of the destruction of lexical variables when they go out of scope, this will also work if the scope is exited by an exception. >> 4) The topic of this ticket: its default set of tests for the directory will >> change (just test if it's a directory?) > Some platform might even be okay with a nondirectory for chdir. This wouldn't > be particularly useful, but we can't assume that all platforms do only useful > things. > > That said, I think the indir tests should just test that chdir is being > called at the right times with the right arguments (can Perl6 tests do > mocking? that would be needed) > So any error case testing can be done for chdir. Since changing the current directory is nothing else than changing $*CWD, mocking should not be a problem. >> Not every change will be possible, because we can't break existing spectests, > Oh, is Perl6 that stable now? > I thought it should still be possible to fix mistakes. I hear they’re still fixing mistakes in Perl 5. Which, in my impression, is considered stable by many :-) Liz
