[PATCH 1 of 4] pull: work as usual bare "hg update" for URL#BRANCH

[FUJIWARA Katsunori]

(CC-ed to Pierre-Yves and Ryan for suggestion from point of view of
DefaultDestinationPlan)

At Sun, 21 Feb 2016 13:17:14 +0900,
Yuya Nishihara wrote:
> 
> On Sun, 21 Feb 2016 03:31:30 +0900, FUJIWARA Katsunori wrote:
> > At Sat, 20 Feb 2016 16:05:32 +0900,
> > Yuya Nishihara wrote:
> > > On Thu, 18 Feb 2016 04:49:25 +0900, FUJIWARA Katsunori wrote:
> > > > # HG changeset patch
> > > > # User FUJIWARA Katsunori <foozy at lares.dti.ne.jp>
> > > > # Date 1455738388 -32400
> > > > #      Thu Feb 18 04:46:28 2016 +0900
> > > > # Node ID 4da9702b2eef28a1c6097f5dbce322bec8925fcd
> > > > # Parent  95bf01b8754016200a99fd3538e78030b2028c60
> > > > pull: work as usual bare "hg update" for URL#BRANCH
> > > > 
> > > > Ordinarily, "hg pull -u" works as same as "hg pull" + bare "hg update"
> > > > (without any explicit revision), for example:
> > > > 
> > > >   - advance current active bookmark, if the destination of the update
> > > >     is valid for 'bookmarks.validdest()'
> > > > 
> > > >   - update not to branch tip but to active bookmark, if the latter is
> > > >     already updated at pulling changes
> > > > 
> > > > But before this patch, "hg pull -u" for "URL#BRANCH" doesn't work as
> > > > expected.
> > > > 
> > > >   - current active bookmark is never advanced
> > > > 
> > > >   - update to branch tip always
> > > > 
> > > > This seems not reasonable for a user, who works on the branch "BRANCH"
> > > > in the repo cloned by "URL#BRANCH".
> > > > 
> > > > But, on the other hand, if "hg pull" is executed with explicit
> > > > --branch/--rev, or "URL#ANOTHER-BRANCH" source, this behavior seems
> > > > reasonable, because user would want to update to specified revision
> > > > (or branch tip).
> > > > 
> > > > This patch treats updating at "hg pull -u" as bare "hg update", if:
> > > > 
> > > >   - no --branch/--rev is specified, and
> > > >   - current branch name is equal to "fragment" of "URL#fragment"
> > > > 
> > > > In this case, 'checkout = None' makes postincoming() use
> > > > destutil.destupdate() to calculate the destination of the update and
> > > > the bookmark to be advanced, as same as usual bare "hg update".
> > > 
> > > New behavior makes more sense to me, but I don't think "pull -u -bBRANCH URL"
> > > should be different from "pull -u URL#BRANCH". Also, it is surprising that
> > > "pull -u URL#BRANCH" does update to BRANCH tip if wdir is not at BRANCH.
> > > 
> > >   $ hg clone URL#A foo
> > >   $ cd foo
> > >   $ hg pull -u  # takes me to max(.::(head() and branch(A)))
> > >   $ hg branch B
> > >   $ hg ci
> > >   ... doing lots of works ...
> > >   $ hg pull -u  # why switching to the branch A?
> > > 
> > > If we go this direction, I think "pull -u" should always select the default
> > > update destination, er, unless -rREV is specified?
> > 
> > Would you mean that both of "-b BRANCH" or "URL#fragment" doesn't
> > affect to determining the destination of the update ?
> 
> Or always affects no matter if wdir is BRANCH or not. It seems confusing for
> me to switch "hg update" and "hg update BRANCH" by dirstate.branch().
> 
>  a) "hg pull -bB -u" == "hg pull -bB && hg update B"
>  b) "hg pull -bB -u" == "hg pull -bB && hg update"
> 
> My opinion is (a) or (b) should be selected statically, not dynamically based
> on current branch.

I think that current updating behavior of "pull -u" below is
reasonable in ordinary case, because user would want to update to
specified one immediately. So, I like (a) above.

  - update to specified revision (or the tipmost branch head),
    if -r REV, -b BRANCH, or URL#X is specified

  - otherwise, update like as bare "update"

A main concern of this patch is:

  even in the repo cloned via "URL#X", "pull -u" treats "#X" not as a
  part of "default" pulling source, but a kind of "-r X"

  => active bookmark isn't advanced as expected

  => branch might be switched unintentionally, as you described in
     previous reply above

Therefore, what about another solution below, even though this is a
kind of "dynamic" one, too ?

  - update like as bare "update", if "[paths] default" and pulling
    source have same fragment part

    (we might have to compare also between "#fragment" and "-b
    BRANCH", for similarity)

  - otherwise, update to specified fragment

Let's ignore the case, in which user changes "[paths] default"
configuration after cloning from "URL#X" :-)


> FWIW, this series would be tightly related to the update destination changes
> which are ongoing. Do you know if "pull --update" was discussed in the last
> Sprint?

No, all I know is the thing described in DefaultDestinationPlan wiki
page. It seems to make no mention of pulling from "URL#fragment".

  https://www.mercurial-scm.org/wiki/DefaultDestinationPlan


----------------------------------------------------------------------
[FUJIWARA Katsunori]                             foozy at lares.dti.ne.jp