Changes between Version 13 and Version 14 of CabalConfigurations

Timestamp:

07/16/07 09:38:42 (6 years ago)

Author:

nominolo

Comment:

added irc log for isaac

CabalConfigurations

@@ -103 +103 @@
 Note that if you want to change the configuration of a package, you have to ''./Setup.lhs clean'' first, to make sure everything gets recompiled.
 
+== Discussions ==
+
+'''2007-07-16'''
+{{{
+16:38 <     Igloo> nominolo: http://urchin.earth.li/~ian/nom.txt
+16:39 <  nominolo> ah. ok. that's pretty much what i agreed on with bringert, too
+16:40 <  nominolo> one thing is that i'm not quite sure what the interface to a cabal file with configs should be
+16:40 <  nominolo> e.g. for cabal install
+16:40 <  nominolo> atm, it's very opaque
+16:41 <  nominolo> ie., you can only read it and find a possible flag assignment
+16:41 <  nominolo> then you get an old-style interface
+16:41 <     Igloo> What can't you do?
+16:42 <     Igloo> Presumably the datastructures are also exported?
+16:42 <  nominolo> inspect which conditions are there, what possibe deps, what archs are possible
+16:43 <     Igloo> But that could always be added later, right? Or written by hand with the datastructure?
+16:43 <  nominolo> atm, it is parsed directly into a structure optimized for flag-resolving
+16:43 <  nominolo> not yet
+16:43 <  nominolo> i think i'll add an intermediate representation
+16:44 <  nominolo> sth. like: data Stmt = CIf Cond [Stmt] [Stmt] | CProp ..
+16:44 <     Igloo> Ah, I see
+16:45 <  nominolo> my internal nstructure is more like a decision tree
+16:45 <  nominolo> or rather, graph
+16:46 <  nominolo> bjorn and i also talked about possible requirements for cabal-install
+16:47 <     Igloo> I'm happy for Bjorn to mentor you rather than me, BTW; might make more sense given the goegraphy
+16:47 <  nominolo> yes, but he'll be away for a while from next week on
+16:47 <     Igloo> Ah, OK
+16:48 <  nominolo> i'll try to summarize it in a mail
+16:49 <  nominolo> btw, what is the problem with building it in HEAD?
+16:50 <     Igloo> The problem is building the HEAD with it; the last part of http://www.haskell.org/pipermail/cabal-devel/2007-June/000652.html
+16:50 < lambdabot> Title: RFC: Cabal-Configurations patch(es), http://tinyurl.com/36yqf7
+16:52 <  nominolo> ah, right.
+16:57 <     Igloo> If it works as much as the old Cabal then I wouldn't be shy about pushing it into the HEAD before it is perfect, by the way
+16:58 <     Igloo> The sooner we can get the HEAD to use configurations the more chance we have of finding any bugs
+16:59 <  nominolo> yes.  i'll try to fix that today
+16:59 <     Igloo> Cool
+17:00 <  nominolo> the other changes shouldn't affect backwards compatibility
+17:54 <  nominolo> Igloo: libraries/installPackage.hs is supposed to register an already configured (and built) package, right?
+17:54 <     Igloo> nominolo: Yes
+17:55 <  nominolo> ok. than the fix is simple, because finalizePackage has already been called
+17:55 <  nominolo> configure writes down a resolved package description.  currently into a separate file
+17:56 <     Igloo> Ah, I thought it just put the flags used into the LBI
+17:56 <  nominolo> i guess it would be a good idea to put that into the persistBuildConfig
+17:57 <  nominolo> i don't think this would help with anything
+17:57 <  nominolo> could be done of course
+17:57 <     Igloo> Hmm, actually, why isn't the whole PackageDescription in the LBI?
+17:57 <  nominolo> yes. that's what i'm considering now
+17:58 <  nominolo> i wonder why we have to read the PD anyways
+17:58 <     Igloo> The reasons I can see for putting the flags in are (a) you can find out what flags were used and (b) tweaks to things like the package description won't need a reconfigure
+17:58 <     Igloo> When installing? So we know what modules we need to install
+17:59 <     Igloo> (and other files, as appropriate)
+17:59 <  nominolo> i see
+18:00 <  nominolo> then we should just use the finalized PD from the persistBuildConfig
+18:02 <  nominolo> that's good.  this way i can undo some interface changes
+18:03 <     Igloo> Oh, I'm expecting it to need more interface changes, as most of the hooks will no longer need to take a PackageDescription argument
+18:03 <     Igloo> OAIC?
+18:04 <  nominolo> true
+18:06 <  nominolo> would we want to detect if a .cabal file was changed after a configuration was called?
+18:06 <  nominolo> ie, if we read both files separately we could do that
+18:06 <     Igloo> I'm actually liking the idea that no changes are noticed, as I think currently it's somewhat random whether one is
+18:06 <  nominolo> but also get inconsistencies
+18:06 <  nominolo> yeah
+18:06 <     Igloo> And yeah, it means the cabal file can't go inconsistent with the config result
+18:07 <     Igloo> So bundling the PD in the LBI makes sense to me
+18:07 <  nominolo> yeah, i think so, too
+18:07 <     Igloo> Then the configure hook gets a PPD and all the other hooks get an LBI, I think
+18:07 <     Igloo> Wait, I've got PD and PPD the wrong way round, haven't I
+18:07 <  nominolo> yes, and i'll change the PPD to something more usable
+18:08 <  nominolo> no it's right that way round
+18:08 <     Igloo> Oh, OK. What's the first P again? I thought it was Prepared
+18:08 <  nominolo> parse -> PPD -> config+finalize -> LBI+PD
+18:08 <  nominolo> prepared
+18:08 <  nominolo> not super-sensible
+18:09 <  nominolo> maybe it should be PD instead of PPD, and LocalPD 
+18:09 <  nominolo> but that might break too much
+18:10 <  nominolo> i'd like to keep the current names afap
+18:11 <     Igloo> current as in before your patches?
+18:11 <  nominolo> yup
+18:11 <  nominolo> i mostly just added code
+18:11              Igloo doesn't like prepared, but can't think of a good word OTTOMH
+18:12 <  nominolo> ParsedPD?
+18:12 <  Philippa> compiled?
+18:12 <     Igloo> I think it should be something like Complete or Flagful
+18:13 <     Igloo> And we can add a type synonym of an equivalent name for PackageDescription, and migrate to it
+18:13 <  nominolo> well, the thing is, i kinda "compile" to the old-style PD
+18:13 <     Igloo> Oh, except we can't do that with the constructor name. Why don't we have constructor synonyms?
+18:14 <     Igloo> I'm not keen on parsed, because you could put that on the front of pretty much any datastructure
+18:14 <  nominolo> yeah, that would be helpful
+18:14 <  nominolo> true
+18:14 <  nominolo> it should be PD and CompiledPD
+18:15 <     Igloo> CompletePD and (Local)PD?
+18:15              Igloo is no good at naming things
+18:16 <  nominolo> GenericPD and (Local)PD?
+18:16 <  nominolo> GlobalGD and (Local)PD
+18:16 <  nominolo> *GlobalPD
+18:17 <     Igloo> Or Generic and Specific, maybe
+
+}}}
+
 
 == Progress Log ==