[xcvb-devel] Why (module... ) must be specified in every file?

[Anton Vodonosov]

I have been thinking about it for several days, but still haven't
understood it completely :)

So maybe you may just say it's possible, but not implemented :)

Best regards,
- Anton

on Sunday, October 25, 2009, 5:14:47 PM Faré wrote:

> 2009/10/24 Anton Vodonosov <avodonosov at yandex.ru>:
>> Why in XCVB (module... ) must be specified in every file?
>>
>> What makes it insufficient to just list the files in the build.xcvb
>> (e.g. compiling them in the order they are listed)? Desire to
>> parallelize the compilation (not total order, in contrast to
>> :depends-on list in build.xcvb)?
>>
>> Is the (module... ) in every file optional? What will happen
>> if I leave only :depends-on list in build.xcvb and remove all
>> the (mudule... ) in separate files?
>>
> That's kind of a FAQ, so I probably should answer it in the README.
> Here's more or less what I'll add to it...


> Why can't module declarations be moved to a central file?
> =========================================================

> The reason why there needs be a module in every file is that
> (at least in the current version of XCVB) is that we build using Make,
> which uses change detection at the file resolution (using timestamps),
> to determine whether or not to recompile object files.

> The failure scenario is what happens when you modify the dependencies
> of a Lisp file in a way that changes its semantics
> (e.g. a missing macro, special-variable or package declaration, etc.,
> may cause an error in one case, not the other).
> If you want reliable deterministic compilation (the goal of XCVB),
> then you want to recompile any time such dependency modification happens.
> If you put all the dependencies in a central file, then
> whenever the central file is modified because a new file
> was added or its dependencies modified,
> then everything will have to be recompiled.
> Or if you may trust modifications to the central file to not matter,
> and experience subtle failures.

> Note that recompiling based on file-contents instead of file timestamp
> would have the same issue. The issue is the resolution of change detection.
> The solution that would allow to reconcile reliable incremental compilation
> with a centralized dependency definition is what in the XCVB TODO file
> I called "Exploded File Dependencies": for the sake of detecting changes,
> explode the centralized dependency file into one file per Lisp module,
> with each per-module exploded bit containing all the dependency information
> about that module, and only that information. It's by no means impossible,
> but it's just something painful and non-trivial that hasn't shown anywhere
> near the top of my TODO list yet.

> [ François-René ÐVB Rideau | Reflection&Cybernethics | http://fare.tunes.org ]
> There is no such thing as philosophy-free science; there is only science
> whose philosophical baggage is taken on board without examination.
>         -- Daniel C. Dennett, Darwin's Dangerous Idea