Hi anyone interested in Sage (the distribution's) build system, I have a ticket that's been languishing for months now which makes some significant overhauls to how Sage's build/make/Makefile is generated and structured: https://trac.sagemath.org/ticket/21524 I know we're all very busy and that it's a difficult patch to evaluate, so I haven't been too pushy about trying to get eyes on it.
But I've been doing some new work (rather successful so far) to implement a general method for how I can easily add configure-time checks to use system packages for more of Sage's dependencies, and I've been doing this work on top of #21524. So if I want to proceed any further with that work I think progress is needed on this ticket. The basic issue is that Sage's old configure.ac contains code in it for directly writing the Makefile, rather than using a template (a Makefile.in) per the usual autoconf pattern. Of course, the fact that we weren't doing that isn't *wrong*--after all there's nothing else very "standard" about Sage's Makefile--so there was some question as to why this was worth doing at all. I think my approach to fixing the issue answers that question: I think the resulting Makefile, and its template, are easier to understand for one. And the resulting configure script runs *much* faster. Especially on Cygwin where it has always been painfully slow, but the speedup is even noticeable on a fast Linux machine. To summarize what this ticket changed: 1) The configure.ac template does not contain code that directly writes the Makefile. Instead, it loops *once* over all the packages in build/pkgs, classifies them into various categories, and generates some variables that are easily consumed by make (e.g. a list of standard packages, and variables containing each package's dependencies). 2) These variables are then inserted into a Makefile template (Makefile.in) by config.status, at the end of running the configure script, via the standard AC_OUTPUT mechanism. 3) Unlike the old Makefile, neither the new Makefile nor the template it is generated from contain explicit rules for building each package. As a result it's actually quite a bit shorter. Instead it uses relatively advanced features of GNU make (but easy to understand, I think, especially with the inline comments I've added)--it uses the variables substituted into the Makefile.in by configure along with some simple templates and foreach/eval loops to generate the build and clean rules for each package. The result is a more compact Makefile that, for the most part, just contains some variables with information about the individual packages, and some templates for generating rules for the packages. Most other parts of the Makefile, such as those included from build/make/deps, remain mostly unchanged for now. The resulting configure script is also much faster, as it only loops over the packages once and performs some relatively simple checks, since it has much less text to generate. The old configure had a shell function in it called filtered_packages_list that would loop over all the packages and filter them according to some criteria, and the old configure script called this four or five times, which was responsible for much of the slowness. The new script loops over the packages only once to generate the variables needed by the Makefile. Hopefully this makes sense, and will be of interest to someone. I'm not 100% married to all the details of my current approach, but I think the general approach is much cleaner and, as I've noted, faster. Best, E -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To post to this group, send email to sage-devel@googlegroups.com. Visit this group at https://groups.google.com/group/sage-devel. For more options, visit https://groups.google.com/d/optout.
