bug#40371: [R7RS] Guile does not accept library name parts that are non-negative exact integers

(import (srfi 9)) doesn't work, but (import (srfi :9)) does.



> On 2 Apr 2020, at 21:47, Marc Nieper-Wißkirchen <[email protected]> wrote:
> 
> 
> 
> Am Do., 2. Apr. 2020 um 21:05 Uhr schrieb Andy Wingo <[email protected] <mailto:[email protected]>>:
> In the concrete case of the SRFI modules, importing e.g. (srfi 9) works
> AFAIU.  Does this not work for you?
> 
> I use Guile 3.9.1.
> 
> I can do (import (srfi srfi-9)), but I can't do (import (srfi 9)).
> 
> That latter yields the error:
> 
> source expression failed to match any pattern in form (srfi 9).
>  
> 
> I think that allowing numbers as module name components, beyond the SRFI
> modules, is not currently a good idea for Guile.  I had a look at it and
> it's a bit too intrusive.
> 
> If numbers are not allowed, Guile will be severely crippled with respect to R7RS code. Most SRFIs are distributed under the name `(srfi NNN)' so many R7RS programs intended to be portable will try to import libraries of the form, say `(srfi 9)' and Guile would complain.
> 
> `cond-expand' is not helpful here in general as an R7RS top-level program has to start with an import and cannot start with some `(cond-expand (guile ...))'. (Besides, `cond-expand' has its own problems: https://lists.gnu.org/archive/html/bug-guile/2020-03/msg00097.html <https://lists.gnu.org/archive/html/bug-guile/2020-03/msg00097.html>).
> 
> As a quick-and-dirty workaround, I would suggest that the Guiles (syntax-case?) parser of library names accepts numbers as module name components but treats them internally as symbols (say, by prefixing them with a colon) so that the main module code doesn't have to be touched. The locator for library code in the file system will then have to look for a filenname with a colon and without.
> 
> Marc
> 
>  
> 
> Andy
> 
> On Wed 01 Apr 2020 12:47, Marc Nieper-Wißkirchen <[email protected] <mailto:[email protected]>> writes:
> 
> > An R7RS library name consists of parts, where each part is either a symbol or
> > a non-negative exact integer. Guile doesn't support the latter ones.
> >
> > This is unfortunate as the implementation of a SRFI NNN is usually delivered
> > in form of a library named (srfi NNN).
> >
> > When this is corrected, for interoperability, it would be great if Guile offers
> > the included SRFIs not only under the name (srfi srfi-NNN) but also under
> > (srfi NNN).
> >
> > Thanks,
> >
> > Marc