std.path review: second update

[Jonathan M Davis]

On Saturday 30 July 2011 15:38:33 Lars T. Kyllingstad wrote:
> On Sat, 30 Jul 2011 23:27:33 +0800, KennyTM~ wrote:
> > On Jul 30, 11 17:00, Lars T. Kyllingstad wrote:
> >> On Sat, 30 Jul 2011 03:47:55 +0800, KennyTM~ wrote:
> >>> - hasDrive, isDriveRoot: the path
> >>> 
> >>>       "#:\x"
> >>> 
> >>> should not pass hasDrive. In Windows only /[a-zA-Z]/ are supported
> >>> drive letters. '#:' may work but is not officially supported.
> >> 
> >> True, but then "#:" must be considered a directory name, which is not
> >> much better.  The module generally assumes that any path you pass to
> >> it
> >> is well-formed.  If not, the results are undefined.
> >> 
> >> The rationale for this is that you don't know whether a path is valid,
> >> i.e. well-formed AND correct (pointing to the right place in the file
> >> system), until you try to use it.  std.path can in principle verify
> >> well- formedness, but since it cannot verify correctness, both may
> >> just
> >> as well be taken care of by the OS.
> > 
> > Fair enough.
> 
> I've added a note stating this to the module documentation.
> 
> >>> - Bug 6390 (= 6337) has been fixed. Some CTFE tests can be enabled.
> >> 
> >> I know, but the fix hasn't been released yet.  I'd prefer if people
> >> didn't have to build the compiler from repo to test the module.  I'll
> >> be sure to enable the tests when DMD 2.055 has been released.
> > 
> > If this module is accepted, it will be released along with 2.055 which
> > carries the fix of 6337 also. Have to delay those tests to 2.056 doesn't
> > sound reasonable to me. It's OK if those are just disabled to ease
> > reviewing, but I'd argue that in the final commit, the tests must be
> > enabled whenever the trunk allows it.
> 
> I think the current policy is that Phobos trunk must compile with the
> latest released compiler.  I'll check up on this, and if I'm wrong, I
> will of course enable the tests when I commit the code to Phobos.

That's wrong. The latest druntime and Phobos in github do _not_ have to 
compile with the latest released version of the compiler - just the latest 
version of the compiler on github. The autotester is constantly running with 
the latest of all 3, and it's not altogether uncommon that a change is made in 
druntime and/or Phobos so that they'll still compile after a change to the 
compiler (or even upon occasion that a change is made in the compiler to make 
something in druntime or Phobos possible). It would be problematic for dmd's 
tests if Phobos and druntime were forced to work with the released version of 
the compiler instead of the latest in github, and it could cause problems for 
druntime and Phobos when you went to do a release. Such issues are likely to 
be fewer and farther between than they used to be, since the language doesn't 
have huge changes from release to release, but both the compiler and the 
libraries are in enough flux that tying the libraries to the last released 
version of the compiler would be a problem.

I don't think that it's unreasonable to comment out some tests during the 
review process so that people aren't forced to grab the latest dmd from 
github, but they should be uncommented if/when the code gets merged into 
Phobos.

- Jonathan M Davis