Fixing CFFI and JAVA:JNEW-RUNTIME-CLASS (was Re: r14858 (was Re: r14857 breaks CFFI (was Re: Assorted patches)))

Mark Evenson evenson at
Sat Oct 15 12:53:10 UTC 2016

On 2016/9/30 00:15, Mark Evenson wrote:


>> Apparently, this [one breaks CFFI][1].  Looking at how to fix, but my
>> time is limited, as I am real busy with the [latest startup to use ABCL][2].
> Incorrect triage on my part:  the [actual offending patch][r14858] is
> the tres cool extension to the generation of synthetic runtime
> constructors contributed by Olof (  The
> patch seems to cause memory corruption (in various NNNstrcmpXXX parts of
> libc) in at least the currently shipping macOS ORCL JVM.  More tests
> pending.


> [r14858]:
>> [1]:

Good news:  thanks to a tip from Anton, I have worked out the source of
my problems with getting Olof's work on JNEW-RUNTIME-CLASS.  The problem
occurs when one is using a pre cffi-0.18.0 release, as is the case with
Quicklisp 2016-09-29.  Presumably, both Olof and Elias are using the
"correct" version of cffi, so never saw these problems.  Note to future
self:  provide a simple Lisp form that would transcribe how all
ASDF/Quicklisp dependencies are resolved on a given installation for
diagnostics in the future

For those wishing to "patch" existing Quicklisp installation one can add
a functioning version of cffi-0.18.0 quickly via:

  cd ~/quicklisp/local-projects && git clone -b v0.18.0

All ASDF definitions that are placed under
<file:~/quicklisp/local-projects> override those defined by a given
Quicklisp dist.  When Quicklisp gets updated, be sure to remove this
local directory.

[Olof's contributions to JNEW-RUNTIME-CLASS][r14903] have been restored
to trunk in all their glory.


"A screaming comes across the sky.  It has happened before, but there
is nothing to compare to it now."

More information about the armedbear-devel mailing list