[bug#53878] [PATCH v5 03/22] gnu: racket: Use Git origins for Racket packages.

[Philip McGrath]

On Saturday, February 26, 2022 1:30:08 PM EST Liliana Marie Prikler wrote:
> Hi,
> 
> Am Samstag, dem 26.02.2022 um 12:23 -0500 schrieb Philip McGrath:
> > Hi,
> > 
> > On Saturday, February 26, 2022 9:26:29 AM EST Liliana Marie Prikler
> > 
> > wrote:
> > > I'm not talking about the generated file, but the procedure that
> > > generates it.  Special characters like dots and slash are still
> > > pretty special in Scheme, even if they're allowed as identifiers.
> > 
> > Do you mean A, B, C, and/or D of these:
> > a
> > 
> > >           (replace 'configure
> > >             (lambda* (#:key inputs configure-flags #:allow-other-
> > > keys)
> > >               (let* ((racket (search-input-file inputs
> > > "bin/racket")))
> > >                 (apply invoke
> > >                        racket
> > >                        #$make-installation-layer.rkt ;; <--- A
> > >                        `(,@configure-flags
> > >                          ,(dirname (dirname racket))
> > >                          ,#$output))
> > >                 (invoke racket
> > >                         "--config" (string-append #$output
> > > "/etc/racket")
> > >                         "-l" "raco" "setup"
> > >                         "--no-user"))))
> > > 
> > > [...]
> > > 
> > > (define make-installation-layer.rkt ;; <--- B
> > >   (scheme-file
> > >    "make-installation-layer.rkt" ;; <--- C
> > >    `(module
> > >      make-installation-layer racket/base ;; <--- D
> > >      (require racket/cmdline
> > 
> > or something else?
> 
> The name in B is confusing imo, because it implies a procedure when it
> is in fact a file.  Of course, that'd affect A as well.  The file name
> in C does not matter too much, but should probably be matched with B to
> some extent (but importantly keeping the .rkt extension even if we drop
> it from B).  I'm not sure if D is a function call or not, but if it's a
> module name perhaps we should make it match B.
> 
> Given that we pass configure-flags to this thing, how about "configure-
> racket-layer", "configure-layer.rkt" or something along those lines?
> 

"configure-layer.rkt" seems fine. I do see the potential for confusion about 
whether this was "making" an "installation-layer.rkt".

(D is a module name and should for clarity match C without the suffix. 
Ultimately, for a module that is not a submodule, it is ignored and the name 
comes from context. This is an older way of writing modules that predates 
"#lang" (which would be inconvenient with `scheme-file`). Calling `read` with 
the appropriate parameters on the input "#lang racket/base 42" would produce 
the datum `(module anonymous-module racket/base (#%module-begin 42))`.)

> > In particular, not to be dense, but B is not a function.
> 
> Yeah, that's my bad for only skimming these and not paying too close
> attention to brackets.  The name itself implies a function, hence my
> confusion.  Looking at things closely would make me see more typos like
> ";; workaroung Guile reader/printer:"
> workaround.
> 

Thanks.

> > > > > I think some way to shorten those origins would do wonders in
> > > > > terms
> > > > > of the number of lines this patch adds.
> > > > 
> > > > At one point, I had abbreviated the origins as something like:
> > > > 
> > > >   (REPO-SPEC SHA256 [COMMIT])
> > > > 
> > > > [...]
> > > 
> > > I wouldn't do this inside of extract-package-source, but define a
> > > one or two liner for adding specifically packages hosted on
> > > racket's github.  WDYT?
> > > 
> > > > I wasn't sure about the trade-off between being slightly more
> > > > cryptic
> > > > than explicit origins, but a lot shorter.
> > > 
> > > Note that the goal is not to code golf, but to be understandable. 
> > > When adding a bunch of origins as inputs, understandability suffers
> > > by induced scrolling.
> > 
> > I've done this, and it was a very good idea. In particular, it helped
> > find some places where I hadn't switched to using `%racket-commit`
> > once the release was tagged.
> 
> That is better, but can still be improved.  As hinted at in my comment
> to your tree, origin snippets might be a better solution here unless
> I'm missing something in the heat of the moment.

Could you give an example of what you imagine the result might look like in 
use? E.g. (but I'm not sure this would really be an improvement):

> (cons*
>  ...
>  libedit
>  racket-minimal
>  (append
>   (racket-package-sources
>    "2d" (base32 "1zzcz5qyjv7syi41vb8jkxjp1rqgj61zbsdrg0nlc4qy9qsafzgr")
>    "2d" "2d-doc" "2d-lib")
>   (racket-package-sources
>    "algol60" (base32 "09kj6asypmc24n29w0izc9p0q8hpga2hpkchsypfwn5c8zpvihlx")
>    '("algol60" "."))
>   ...))

-Philip

signature.asc
Description: This is a digitally signed message part.