SWANK breaks trying to read SB-DISASSEM::ALIGN due to a package lock

[Paul F. Dietz]

Was SB-DISASSEM locked recently, or did this symbol go away?

EDIT: I see it was renamed ALIGN-UP. SWANK will need to be updated. Also, why was SWANK referencing a non-exported symbol? API issue here.

[Paul F. Dietz (paul-f-dietz)]

Renaming ALIGN to ALIGN-UP in swank does not fix this.

[Douglas Katzman (dougk)]

cripes. ALIGN was never supposed to be for users (including swank)
align-up is declared inline, so it's inlined and the tree shaker removes it.
Even if it were renamed back to ALIGN it would have that issue. This is sad.

That said, going back to the initial commit in which swank/sbcl started to call disassemble-code-component, it's bogus (certainly now, probably before).

commit 70d4dc9100d3330561f8be12f917822b4a3ab5e4
Author: Helmut Eller <email address hidden>
Date: Wed Nov 24 19:58:37 2004 +0000

 (cond ((sb-kernel:%code-debug-info o)
        (sb-disassem:disassemble-code-component o :stream s))
       (t
        (sb-disassem:disassemble-memory
         (sb-disassem::align
          (+ (logandc2 (sb-kernel:get-lisp-obj-address o)
                       sb-vm:lowtag-mask)
             (* sb-vm:code-constants-offset
                sb-vm:n-word-bytes))
          (ash 1 sb-vm:n-lowtag-bits))
         (ash (sb-kernel:%code-code-size o) sb-vm:word-shift)
         :stream s)))))))

It SHOULD NOT be trying to compute the bounds on what to disassemble.
It should just call disassemble-code-component in any case. (The COND should go away)
The second branch became even *more* wrong after Stas changed SB-KERNEL:%CODE-CODE-SIZE to return bytes instead of words. (The ASH shouldn't be there)

So fundamentally, this is all wrong and somebody with Slime commit access should simplify that code

[Paul F. Dietz (paul-f-dietz)]

I created an issue for this over in the Slime github.

https://github.com/slime/slime/issues/429

[Stas Boukarev (stassats)]

In ef3f1195c7ea8fc81b8d526b9a0d0425cc507e7f.