[Buildbot-devel] Few things I'd like ...

[Greg Ward]

[Frédéric posts a patch for running autogen]
> I joined a patch for this one. It works, but review is welcome.

OK. Better late than never.

@@ -54,11 +54,15 @@ class GNUAutoconf(BuildFactory):
     def __init__(self, source, configure="./configure",
                  configureEnv={},
                  configureFlags=[],
+                  autogen=None,
                  compile=["make", "all"],
                  test=["make", "check"]):
         assert isinstance(source, tuple)
         assert issubclass(source[0], BuildStep)
         BuildFactory.__init__(self, [source])
+        # for some project, we need to generate configure
+        if autogen is not None:
+            self.addStep(Autogen, command=autogen)
[...]
+class Autogen(ShellCommand):
+
+    name = "autogen"
+    haltOnFailure = 1
+    description = ["autogen..."]
+    descriptionDone = ["autogen"]
+    command = ["./autogen.sh"]

Two comments:

1) isn't having a default 'command' attribute in Autogen
semi-pointless, since you always instantiate it with a 'command'
argument?  (OTOH, I suppose this makes it easier for people outside of
GNUAutoconf.__init__() to instantiate Autogen.)

2) wouldn't it be more powerful and general if the 'autogen' argument
to GNUAutoconf.__init__()  were a BuildStep object rather than a
command?  I don't know precisely how you use it now, but I assume it's
something like

  factory = GNUAutoconf(..., autogen=["./autogen.sh"], ...)

which pretty much limits you to what your Autogen build step can do.
But if you were allowed to pass any BuildStep object, that call would
become

  factory = GNUAutoconf(..., autogen=Autogen(), ...)

which is less typing, less specific knowledge about how to invoke
autogen.sh, and more general.  And it makes my comment #1 irrelevant,
since now the default Autogen.command is very sensible and useful.

Greg

P.S. I don't see a ticket for this in the buildbot trac system.