[asdf-devel] I think new ASDF has busted asdf-binary-locations

[Faré]

Note that I have been maintaining my variant of A-B-L as part of cl-launch,
and upon noticing that resolve-symlinks had been removed, I immediately
issued a new release 2.23 of cl-launch.

I for one would welcome A-B-L becoming part of ASDF, so I could have
cl-launch just use that (after a transition period).

I agree that logical pathnames are indeed broken as specified, and
that since at least SBCL enforces the specification, they shouldn't be
used.

[ François-René ÐVB Rideau | Reflection&Cybernethics | http://fare.tunes.org ]
To stay young requires unceasing cultivation of the ability to
unlearn old falsehoods. -- Robert Heinlein, "Time Enough For Love"


2009/9/9 Robert Goldman <rpgoldman at sift.info>:
> james anderson wrote:
>> hello;
>>
>> i recall, that we have started down this path before, but we never
>> got very far, so i would like to pick up the thread again:
>>
>> what exactly fails (or is just inconsistent) in the respective
>> logical pathname implementations to preclude accomplishing the same
>> thing with logical pathnames?
>> whatever that may be, why is it better to re-implement the
>> functionality for asdf rather than to fix the problem?
>
> Two things rule out logical pathnames as a solution for this problem:
>
> 1.  They don't actually solve the problem --- in order to get what you
> want you'd still need to have logic that redirects the binary files to
> different directories.  I.e., we'd have to add logic to differentially
> define logical pathnames for binaries depending on features of the lisp
> implementation and then we'd have to add logic to methods for
> output-files.  Gary's A-B-L just fixes this with methods for
> output-files.  So it's a simpler solution and more portable.
>
> 2.  Logical pathnames are defined in ANSI CL to use case-flattened
> pathnames.  That means they are an extremely poor fit for modern
> case-sensitive file systems.  Some number of existing ASDF systems would
> break because their directory structures contain case-sensitive
> pathnames.  From the Hyperspec grammar for logical pathname namestrings
> (section 19.3.1):
>
> "word---one or more uppercase letters, digits, and hyphens."
>
> As long as SBCL hews to the letter of the ANSI spec for logical
> pathnames, I regard logical pathnames as useless in portable code.  I
> now use them only in code that, for one reason or another, will only run
> in ACL.  [Note that this is /not/ meant as a criticism of the SBCL policy.]
>
>
>
>
> So Gary's existing solution is (a) more portable; (b) simpler; and (c)
> less damaging to existing ASDF systems; (d) it's done and (e) it's
> extensively tested in the wild.
>
> I think the only question is whether we should make A-B-L optional as
> now (but distribute it with stock ASDF) or, as Attila suggests,
> integrate it fully, but configure it so that it behaves as current stock
> ASDF.
>
> Best,
> R
>
>
>>
>> On 2009-09-09, at 14:39 , Robert Goldman wrote:
>>
>>> Gary King wrote:
>>>> (cc'd to list)
>>>>
>>>> Damn. The function disappeared recently (by my hand). I didn't
>>>> realize
>>>> (obviously) that it was used. I'll fix.
>>> Gary,
>>>
>>> Maybe this would be a good time to push A-B-L into the ASDF
>>> repository?
>>>  I've always been in favor of this, acnyway, since it's such a
>>> critical
>>> extension.  IMO it would be great if anyone who had ASDF could also
>>> get
>>> A-B-L with no more work than a call to asdf:oos.
>>>
>>> Best,
>>> r
>>>
>>>> On Sep 8, 2009, at 11:17 PM, Robert Goldman wrote:
>>>>
>>>>> I just updated, and now can't start up lisp because
>>>>> asdf-binary-locations calls ASDF::RESOLVE-SYMLINKS which seems to
>>>>> have
>>>>> vanished from asdf.lisp.
>>>>>
>>>>> Any insight?