Nikita Karetnikov <[email protected]> skribis: > I apologize for the delay. What do you think about the attached patch?
No problem, thanks for working on this! > I noticed that 'guile-gnome' has two 'doc-url' fields. How can I handle > this? (I ignored it for now.) I think it’s a mistake in this case (could you report it to Karl?). It seems safe to ignore. > Also, is there a better way to create 'gnu-package-descriptor'? Note > that I don't want to use setters. Yes, use the wonderful ‘define-record-type*’ from (guix utils). This is what we use for ‘package’, etc. See <http://lists.gnu.org/archive/html/guile-user/2012-11/msg00016.html>, for details. >> This code is run by the user’s Guile, which may be older than 2.0.7 >> (‘http-get*’ was introduced in 2.0.7), so you can’t rely on it. > >> What you can do is something along the lines of what (guix build >> download) does, but always return a port. Maybe there’s a way to share >> code. > > But how can I return a port with 'http-get'? On Guile < 2.0.7, you’ll get a string, so you can just call ‘open-input-string’ to wrap it in a port. On later versions, you’ll get a port. >> I’d rather change ‘group-packages’ to ‘read-package-fields’ or something >> like that. > > I changed it to 'group-package-fields' and added some comments. > >> identity > > (identity sublst) won't work. > > If (regexp-exec package-line-rx (first sublst)) returns #t, > 'and=>' will call 'identity' with the result of 'regexp-exec'. But it > should return 'sublst' instead. Ah, you want (and=> (regexp-exec ...) (const sublst)). But then, it’s equivalent to just (and (regexp-exec ...) sublst). > +(define-record-type <gnu-package-descriptor> Yes, define-record-type* will definitely help. > + (package get-gnu-package-name) Here the field name should rather be ‘name’, for consistency. Never ever write ‘get-’ for getters. ;-) It provides no useful information. See <package> and other records, as examples. > (define (official-gnu-packages) > "Return a list of GNU packages." This should return directly a list of <gnu-package-descriptor> IMO, so ‘create-gnu-package-descriptor’ would be moved here. > + (define (group-package-fields port state) > + ;; Return a list of lists where /most/ inner lists are the GNU > + ;; packages. Note that some lists are not packages at all; they > + ;; contain additional information. So it is necessary to filter > + ;; the output. > + (let ((line (read-line port))) > + (define (match-field str) > + ;; Packages are separated by empty strings. Each package is > + ;; represented as a list. If STR is an empty string, create a new > + ;; list to store fields of a different package. Otherwise, add STR > to > + ;; the same list. > + (match str > + ('"" Extra quote. > + (group-package-fields port (cons '() state))) > + (str > + (group-package-fields port (cons (cons str (first state)) > + (drop state 1)))))) > + > + (if (eof-object? line) > + (remove null-list? state) > + (match-field line)))) > + > + (reverse (map reverse > + (group-package-fields (http-fetch* %package-list-url) > + '(()))))) Apparently ‘create-gnu-package-descriptor’ doesn’t rely on the order of fields, so (map reverse ...) can be omitted, no? > +(define (find-packages regexp) > + "Find packages that match REGEXP." > + (define (create-gnu-package-descriptor package) s/create-gnu-package-descriptor/alist->package-descriptor/ ? With a comment saying that PACKAGE is an alist. > network to check in GNU's database." > ;; TODO: Find a way to determine that a package is non-GNU without going > ;; through the network. > - (let ((url (and=> (package-source package) origin-uri))) > + (let ((url (and=> (package-source package) origin-uri)) > + (pname (package-name package))) s/pname/name/ > (or (and (string? url) (string-prefix? "mirror://gnu" url)) > - (and (member (package-name package) (official-gnu-packages)) I think the last expression will become: (find (compose (cut equal? name <>) package-name) (official-gnu-packages)) Thanks! Ludo’.
