Boost logo

Boost-Build :

From: Jurko Gospodnetiæ (jurko.gospodnetic_at_[hidden])
Date: 2008-04-30 15:38:30


   Hi all.

   generators.jam module seems to have traces of comments and
implementation related to special handling or some multiple target
generation cases which seems to have been implemented at one time but
then incompletely removed in revision 29361 (committed by Volodya).

   I'm attaching a patch (description: Removed some old corpse comments
and debugging output code related to functionality long removed in
revision 29361.) based on revision 44943 and if no one objects will
commit it.

   Volodya, could you look at the following paragraph in its main module
comment and say if it is OK to remove it? Or should it be transformed
into something else? It seems outdated as the construct() rule no longer
takes any such parameter.
-------------
   That early conversion is not always desirable. Suppose a generator
got a source of type Y and must consume one target of type X_1 and one
target of type X_2. When converting Y to X_1 extra target of type Y_2 is
created. We should not try to convert it to type X_1, because if we do
so, the generator will get two targets of type X_1, and will be at loss
as to which one to use. Because of that, the 'construct' rule has a
parameter, telling if multiple targets can be returned. If the parameter
is false, conversion of extra targets is not performed.
-------------

   Also, could you look at the paragraph above it which says '(Later
I'll document the rationale for trying extra target conversion at that
point)'? I'd fix if but I am still not at home with how this all
works... :-)

   generator.run() rule has the following debug output line in it:

         generators.dout [ indent ] " multiple:" $(mutliple) ;

   which can never do anything useful both because of the typo in the
variable name and because no variable with even a similar name exists in
this module.

   construct() rule does not take any 'multiple' parameter but its
comment states:
-------------
   If 'multiple' is true, the rule is allowed to return several targets
of 'target-type'.
-------------

   Best regards,
     Jurko Gospodnetiæ

Index: tools/build/v2/build/generators.jam
===================================================================
--- tools/build/v2/build/generators.jam (revision 44943)
+++ tools/build/v2/build/generators.jam (working copy)
@@ -36,15 +36,6 @@
 # attempts that as early as possible. Specifically, this is done after invoking
 # each generator. (Later I'll document the rationale for trying extra target
 # conversion at that point).
-#
-# That early conversion is not always desirable. Suppose a generator got a
-# source of type Y and must consume one target of type X_1 and one target of
-# type X_2. When converting Y to X_1 extra target of type Y_2 is created. We
-# should not try to convert it to type X_1, because if we do so, the generator
-# will get two targets of type X_1, and will be at loss as to which one to use.
-# Because of that, the 'construct' rule has a parameter, telling if multiple
-# targets can be returned. If the parameter is false, conversion of extra
-# targets is not performed.
 
 import "class" : new ;
 import errors ;
@@ -293,7 +284,6 @@
             )
     {
         generators.dout [ indent ] " ** generator" $(self.id) ;
- generators.dout [ indent ] " multiple:" $(mutliple) ;
         generators.dout [ indent ] " composing:" $(self.composing) ;
 
         if ! $(self.composing) && $(sources[2]) && $(self.source-types[2])
@@ -1163,8 +1153,7 @@
 
 # Attempts to create a target of 'target-type' with 'properties' from 'sources'.
 # The 'sources' are treated as a collection of *possible* ingridients -- i.e. it
-# is not required to consume them all. If 'multiple' is true, the rule is
-# allowed to return several targets of 'target-type'.
+# is not required to consume them all.
 #
 # Returns a list of targets. When this invocation is first instance of
 # 'construct' in stack, returns only targets of requested 'target-type',


Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk