not ui object passed to extsetup()

[Martin Geisler]

TK Soh <teekaysoh at gmail.com> writes:

> I noticed that unlike the uisetup() callback, extsetup() doesn't have
> a ui object passed into it, forcing extensions like color have to
> 'capture' the ui via uisetup() and use it later in extsetup(). Can
> anyone share some background on this, please? Thanks.

The extsetup callback was introduced in 863e237b58fb, which came from
this thread:

  http://markmail.org/message/pr6bwayg6pmy7q6x

There are no comments in the thread -- I guess nobody thought of passing
ui when calling extsetup.

This part of Mercurial could benefit from a major overhaul to make it
more regular. Dividing the initialization into well-defined phases would
be a good start. Today, a hgrc file with

  [extensions]
  foo =
  bar =

will be loaded with a call tree like this:

  dispatch._dispatch

    extensions.loadall
      extensions.load
        # add foo module to extensions._extensions
        foo.uisetup(ui)
      extensions.load
        # add bar module to extensions._extensions
        bar.uisetup(ui)

    foo.extsetup()
    commands.table.update(foo.cmdtable)
    bar.extsetup()
    commands.table.update(bar.cmdtable)

    foo.reposetup(ui, repo)
    bar.reposetup(ui, repo)

A problem with the above call tree is that extensions are initialized
while being loaded, instead of doing it in phases:

  dispatch._dispatch

    extensions.loadall
      # add foo module to extensions._extensions
      # add bar module to extensions._extensions

    extensions.uisetupall
      bar.uisetup(ui)
      foo.uisetup(ui)

    foo.extsetup()
    bar.extsetup()

    commands.table.update(foo.cmdtable)
    commands.table.update(bar.cmdtable)

    foo.reposetup(ui, repo)
    bar.reposetup(ui, repo)

If things were grouped like this, an extension like color could do

  mq = extensions.find('mq')
  record = extensions.find('record')
  if mq and record:
    # add color to qrecord

in extsetup without having to think about the load order, and knowing
that the updated qrecord cmdtable entry will be copied correctly into
commands.table.

I don't know if this covers all the corner cases -- probably not :-)

But I think introducing a new set of callbacks and deprecating the old
ones could help streamline this API.

-- 
Martin Geisler

VIFF (Virtual Ideal Functionality Framework) brings easy and efficient
SMPC (Secure Multiparty Computation) to Python. See: http://viff.dk/.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://selenic.com/pipermail/mercurial-devel/attachments/20090811/4f3719a7/attachment.pgp