[Chicken-janitors] #381: Report the top-level binding that an error occurs in

[Chicken Trac]

#381: Report the top-level binding that an error occurs in
-------------------------+--------------------------------------------------
 Reporter:  alaric       |       Owner:       
     Type:  enhancement  |      Status:  new  
 Priority:  major        |   Milestone:  4.7.0
Component:  expander     |     Version:  4.5.0
 Keywords:               |  
-------------------------+--------------------------------------------------
 One bane of my life is errors like this:

 {{{
 Warning: reference to possibly unbound identifier: archive

 Error: module unresolved: ugarit-core
 }}}

 ...when just about *every* procedure in ugarit-core has an 'archive'
 parameter. I eventually traced down the problem with a binary chop by
 putting deliberate references to unbound identifiers in and looking at the
 order of the unbound identifiers reported.

 I'm not sure if line-number information persists into the expander, but I
 think that it should be possible to work out the name of the top-level
 binding (DEFINE, DEFINE-SYNTAX, or whatever) it's inside. I'm guessing
 that the unbound references are detected after any macros that expand
 *into* top-level bindings are expanded, so this mechanism would still
 produce useful results in such situations.

 Inside the expander, ##sys#register-undefined registers undefined symbols,
 but it
 only seems to be called by ##sys#alias-global-hook which is in turn
 called only by ##sys#strip-syntax, and I'm not sufficiently versed in
 Chicken internals to work out what's what above that, as it's called in
 lots of places.

 I see that it's called directly inside many of the core macro
 definitions inside expander.scm; would it therefore suffice to define a
 parameter for the "current top-level definition" and set it to the
 defined name in the macro expanders for DEFINE, DEFINE-SYNTAX, and any
 others, and if that parameter is not at its default value of #f, report
 it in errors (and make the unbound identifier list for the module be a
 list of pairs, mapping unbound identifiers to the top-level definitions
 they're in?

 Would that work, or is strip-syntax called in other dynamic contexts?

-- 
Ticket URL: <http://bugs.call-cc.org/ticket/381>
Chicken Scheme <http://www.call-with-current-continuation.org/>
Chicken Scheme is a compiler for the Scheme programming language.