I'm able to replicate the crash in v5.3.6, too. I need a little more than the lines below, but
#lang scribble/manual @(require scribble/eval) @(void (make-base-eval)) @((make-base-eval) `(require describe)) is enough. Also, I can get a crash when installing just "describe". The problem goes away if I strip `math/bigfloat` our of `describe`. The crash is custodian-related according to a low-level stack trace, the `bigfloat` library uses `register-custodian-shutdown`, etc... see Racket commit 7e42ee2003. That timing is consistent with a bug in v5.3.x fixed in v6.x. I can't think of a simple workaround to avoid the crash, but running `raco setup -j 1` might clean up after the crash. Using `-j 1` avoids building in parallel, since places seem to aggravate the bug. At Fri, 30 May 2014 19:31:32 -0700, Matthew Butterick wrote: > Looking for clues that would help debug the following error — > > I'm converting a package [1] to be multi-collection & thus compatible with > Racket 5.3.5. > > The revised package works with 6.0. But under 5.3.5, I get a fatal > 'segmentation fault: 11' error while building the docs. I by trial & error I > have traced the offending code to the call to `make-base-eval` shown below: > > #lang scribble/manual > @(require scribble/eval) > @(define my-eval (make-base-eval)) > > The segfault occurs when I install the package from the local directory, or > when I use the Github URL for the source. [2] > > Comparing the docs between 5.3.5 and 6.0, I don't see any changes to > make-base-eval, so at this point, I'm stuck for ideas. > > > > > [1] http://github.com/mbutterick/sugar/tree/dev-53 > > [2] github://github.com/mbutterick/sugar/dev-53____________________ > Racket Users list: > http://lists.racket-lang.org/users ____________________ Racket Users list: http://lists.racket-lang.org/users