Regression in 2.007: newly (?) exported symbol ASDF:APPENDF clashes with alexandria:appendf

[Zach Beane]

To reproduce:

  - load asdf 2.007

  - load alexandria

  - (make-package "FOO" :use '("ASDF" "ALEXANDRIA"))

This affects, to some degree or another, at least these projects:

clonsigna
cl-redis
clws
cl-zmq
hemlock
iolib
sw-http
symbolicweb
ucw

[Zach Beane (xach)]

debugger invoked on a NAME-CONFLICT in thread #<THREAD "initial thread" RUNNING
                                                {1002AAEED1}>:
  USE-PACKAGE #<PACKAGE "ASDF"> causes name-conflicts in #<PACKAGE "FOO">
  between the following symbols:
    ASDF-UTILITIES:APPENDF, ALEXANDRIA.0.DEV:APPENDF
See also:
  The ANSI Standard, Section 11.1.1.2.5

Type HELP for debugger help, or (SB-EXT:QUIT) to exit from SBCL.

restarts (invokable by number or by possibly-abbreviated name):
  0: [RESOLVE-CONFLICT] Resolve conflict.
  1: [ABORT ] Exit debugger, returning to top level.

(NAME-CONFLICT #<PACKAGE "FOO"> USE-PACKAGE #<PACKAGE "ASDF">)[:EXTERNAL]
0]

[Faré (fahree)]

Who is using both alexandria and asdf at the same time? :-/

What is the proper solution for this bug?

Unhappily, as I see it, the current solution would be to either
1- stop including asdf in the package use list, or
2- (:shadowing-import-from :alexandria #:appendf)

[Zach Beane (xach)]

Why export this stuff from asdf in the first place? ASDF need not be a utility library.

[Zach Beane (xach)]

Anyway, it looks like most of the trouble is from iolib.base - if that is fixed, I think that will fix all the others too.

[Faré (fahree)]

Some of the utilities are quite useful to a lot of ASDF clients that have to deal with configuration: notably, merge-pathnames* or getenv or split.

Which repo are you using for iolib?

[Zach Beane (xach)]

Is there a way to offer those functions in a way that won't clash with
legacy code that can't be as easily updated as iolib?

I'm using iolib from git://gitorious.org/iolib/iolib.git.

[Faré (fahree)]

The joke is that:

* Towards the end of ASDF 1, Gary King created an ASDF-EXTENSIONS package, that imported (but failed to export) some utilities.
* When I released ASDF 2, I had a different, incompatible set of utilities, that I exported as ASDF-UTILITIES.
* To make bootstrapping easier, I created a package ASDF-BOOTSTRAP in which the initial update-friendly symbol/package-wrangling happens.
* Recently, Sam Steingold (of CLISP) complained that there were too many packages in ASDF, and since no one was using it, I merged ASDF-UTILITIES back into ASDF, and eliminated ASDF-BOOTSTRAP.

So we've come full circle.

In any case, I believe it is right for ASDF to export its essential utilities.
I also believe CL has a namespace management problem, but oh well.

Sorry for breaking things, but one way or the other, the ASDF API *will* change, and it is foolish to randomly (:use :asdf) and expect to never have to adapt to a modified API. If you want a fixed set of symbols, use (:import-from ...) and/or (:shadowing-import-from ...), or don't import and use fully qualified symbol names.

[Faré (fahree)]

I'm re-filing this as a bug against iolib.

For reference the "don't create many packages" bug was
https://bugs.launchpad.net/asdf/+bug/620579

[Zach Beane (xach)]

What about making another project that does nothing but re-export what
you consider to be the essential utilities of asdf? Then people who do
not expect asdf to be a general utility library will not be affected,
and people who want asdf to be a general utility library can access its
utilities by an explicit decision.

[Attila Lendvai (attila-lendvai)]

...maybe you should have ignored Sam Steingold's request... :)

unless he had an objective reason why and how 3 asdf packages are too many.

but it's only my 0.02

[Stelian Ionescu (sionescu)]

Zach, could you please test iolib HEAD(a7fb2e60a65fa3c63975687bf528fa277be4f018) ? That commit should avoid this symbol clash

[Zach Beane (xach)]

iolib HEAD works for me.

Turns out this still breaks SymbolicWeb, though.

[Zach Beane (xach)]

Scratch that re: symbolicweb.