On Mon, Apr 13, 2009 at 2:31 PM, Stephen C. Gilardi <squee...@mac.com>wrote:
> On Apr 13, 2009, at 4:58 PM, bradford cross wrote: >> >> I have written up my journeys so far: >> http://bradfordcross.blogspot.com/2009/04/clojure-build-and-dependency-management.html >> > > Thanks very much for the write-up! > > Given the recent discussions and interest in dependency management I > thought I'd mention an idea in that area that I think is a good one. > > At ILC '09, Francois-Rene Rideau made a presentation about "xcvb": > > http://common-lisp.net/project/xcvb/ > > which is intended to be a successor to Common Lisp's ASDF with some nice > features. > > One of the issues it addresses is the possibly problematic behavior of a > "same world" compiler as we have in Clojure AOT: build dependencies which > are intended to be explicit and correct can be incorrect, but accidentally > work. > > Here's an example: > > A depends on B and C > E depends on A, B, and D, but only declares A and D > > Compile A > Compile E > > works > > Now change A so it no longer depends on B. > > Now compiling E "breaks". (And in the general case, the guy who made > the change to A knows nothing about E.) Yes, I have had nasty issues with this in the past. IMO, explicitly declaring deps and not relying on transitive resolution is the only way to go. > > > In xcvb, the strategy they use to avoid that and make sure all dependencies > are correctly specified is to make the compilation of each file start with > the compiler's "image" in a known, minimal state. They break the build into > a dependency discovery phase and a make-like minimal rebuild phase (which > may actually use "make" itself) to accomplish this. The tradeoff in overall > "system" compilation time is the reloading of compiled code every time it's > needed rather than just once. > > For us, the corresponding strategy would be that the compilation of each > namespace should start with only clojure.core loaded. > > We could adopt this strategy in a given Clojure build system without any > changes to how Clojure works (by launching a new instance of Clojure for > each compilation) which makes it particularly easy to experiment with. I believe we can achieve this via ant and ivy but i am not sure if they can force explicit deps. I'm not sure if a separate closure process is required, I have seen static analysis tools that can do the graph traversal and warn you about transitive dependency resolutions (or unused dependencies.) > > > --Steve > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en -~----------~----~----~----~------~----~------~--~---
