Re: [PATCH] Create INDEX.LIST; Was: Fix references in jar manifests

[Chris Marusich <cmmarusich@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1614 bytes --]

Hi Danny,

Thank you for writing a patch!  It looks good to me.  We will need to
apply this to the core-updates branch, right?  I think that changes to
the ant-build-system will cause all packages that use it to be rebuilt.

Danny Milosavljevic <dannym@scratchpost.org> writes:

> +(define* (generate-jar-indices #:key outputs #:allow-other-keys)
> +  "Generate file \"META-INF/INDEX.LIST\".  This file does not use word wraps
> +and is preferred over \"META-INF/MAINFEST.MF\", which does use word wraps,
> +by Java when resolving dependencies.  So we make sure to create it so that
> +grafting works."

Is that the only reason?  My understanding is that we want to generate
the JAR indices not only (1) to ensure that grafting will work properly,
but also (2) to ensure that the reference scanner will find all the
store paths produced in the output JAR.  Goal (2) is important because
if a store path is used at run-time by the installed software, but the
path is treated as "dead" by the garbage collector, then the path may be
removed during garbage collection, which would break the software.  If
you agree, then could you update the docstring with a little more info?

> + (define (generate-index jar)
> +    (invoke "jar" "-i" jar))
> +  (every (match-lambda
> +           ((output . directory)
> +            (every generate-index (find-files directory "\\.jar$"))))
> +         outputs))

Does the jar program find the classes here via the CLASSPATH environment
variable, which was set earlier during the configure phase by the
generate-classpath procedure?

-- 
Chris

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]