[yocto] Perforce fetcher ignores module and label

[Andrew Bradford]

Hi Katu,

On 08/28 08:56, Katu Txakur wrote:
> 2017-08-11 13:52 GMT+01:00 Andrew Bradford <andrew at bradfordembedded.com>:
> On 08/11 11:34, Katu Txakur wrote:
> > > 2017-08-07 12:17 GMT+01:00 Andrew Bradford <andrew at bradfordembedded.com
> > >:
> > > > On 08/02 10:28, Paul Eggleton wrote:
> > > > > On Wednesday, 2 August 2017 9:40:10 AM CEST Katu Txakur wrote:
> > > > > > I'm still having problems fetching from Perforce. Is there any
> > > > > > documentation based on the latest implementation of the
> > > > > > poky/bitbake/lib/bb/fetch2/perforce.py file?
> > > >
> > > > There's some documentation now in the official Yocto Project
> > > > documentation location for bitbake fetchers [1].  Prior to my changes
> > > > there was very little if any documentation, afaik.
> > > >
> > > > [1]: http://www.yoctoproject.org/docs/2.3.1/bitbake-user-
> > > > manual/bitbake-user-manual.html#perforce-fetcher
> > > >
> > > > > > 2017-07-25 10:05 GMT+01:00 Katu Txakur <katutxakurra at gmail.com>:
> > > > > > > I'm upgrading a recipe that fetches the source code from
> > Perforce.
> > > > > > >
> > > > > > > The old recipe was:
> > > > > > >
> > > > > > > SRC_URI = " \
> > > > > > >   p4://${P4USER}:${P4PASSWD}:${P4HOST}:${P4PORT}@Depot/path/pe
> > > > > > > rforce/...;module=local/path/relativeto/p4;label=${P4CHANGELIST}
> > \
> > > > > > >   "
> > > > > > >
> > > > > > > With the new version of /lib/bb/fetch2/perforce.py, I had to set
> > > > P4PORT
> > > > > > > outside SRC_URI, leaving the recipe with:
> > > > > > >
> > > > > > > SRC_URI = " \
> > > > > > >   p4://${P4USER}:${P4PASSWD}@Depot/path/perforce/...;module=
> > > > > > > local/path/relativeto/p4;label=${P4CHANGELIST} \
> > > > > > >   "
> > > > > > >
> > > > > > > The Perforce fetcher kind of works, but it seems to be ignoring
> > the
> > > > > > > "module" and the "label" attributes. After reading the Python
> > script
> > > > I can
> > > > > > > see that after the "@", only the substring before the first ";"
> > is
> > > > used.
> > > >
> > > > Sorry for not making the label change more clear, but you should be
> > able
> > > > to simply set SRCREV="labelname" and have the functionality you desire.
> > > > However, we don't use labels very often internally, so the use of
> > labels
> > > > in SRCREV isn't tested that often by me.
> > > >
> > >
> > > I don't use labels that much either, however, I used to be able to write
> > > the changeslist number as a label and it would
> > > fetch the code for that changelist. This doesn't work using SRCREV =
> > > "changelistnumber"
> >
> > This does work for me using oe-core morty branch and bitbake
> > 1.32 which is the versions where these changes were first applied and
> > where I did my testing.
> >
> > Like I have a recipe which I put:
> >
> > SRCREV = "184127"
> >
> > The p4 fetcher will go fetch changelist 184127 of the SRC_URI path in
> > perforce.
> >
> > Or if I put:
> >
> > SRCREV = "${AUTOREV}"
> >
> > Then it'll fetch the HEAD/tip of the SRC_URI path in perforce.
> >
> > Is this not working the same way with poky now?
> >
> 
> I'm using yocto pyro with bitbake 1.34.0. When I first fetch the code for
> my recipe, it gets the latest changelist.
> I submit a change and the next time, the recipe thinks there nothing new to
> fetch. Also, using SRCREV = "184127"
> doesn't work for me, it always gets the latest changelist.

Can you post the recipe where this is failing for you?

I could then try to recreate this issue in a sandbox depot we have in
our perforce server and see what's going on.

> >
> > > > This change to using labels, changelist numbers, and allowing the use
> > of
> > > > AUTOREV was made so that the perforce fetcher would act more like the
> > > > other source code control system fetchers.  Prior to my changes to the
> > > > p4 fetcher, the way that recipes using p4 had to be written seemed
> > > > rather awkward to me.  The way it is now isn't perfect, but it's closer
> > > > to how the other fetchers act, I think (and hope).
> > > >
> > >
> > > > > > Is there a way to set module and label/changelist? I have several
> > > > folders
> > > > > > > per recipe that I need to map to different subfolders and with
> > the
> > > > current
> > > > > > > script some of the folders don't get fetched.
> > > >
> > > > I'm not sure I understand enough about how you use perforce and what
> > you
> > > > mean by mapping different subfolders.  Can you explain this in a more
> > > > expanded way to me?  Maybe with some examples?
> > > >
> > >
> > > What I mean by this is taking folders in different locations in Perforce
> > > and fetch them to a custom folder structure inside the {WORKDIR}/p4
> > folder.
> > > This is an example of how I was doing this with the old version of the
> > > fetcher:
> > >
> > > #leave blank, "", for latest revision
> > > P4CHANGELIST = "${PR}"
> > > S = "${WORKDIR}"
> > > SRC_URI = " \
> > > p4://${P4USER}:${P4PASSWD}:${P4HOST}:${P4PORT}@Depot/project
> > 1/folder1/...;module=subfolderunderp4/src/;label=${P4CHANGELIST}
> > > \
> > > p4://${P4USER}:${P4PASSWD}:${P4HOST}:${P4PORT}@Depot/project
> > 2/folder3/...;module=subfolderunderp4/pkg/;label=${P4CHANGELIST}
> > > \
> > > "
> > >
> > > Where folder1 contains the src files and folder3 contains the pkg
> > > information. These need to be in an specific folder structure for the
> > > recipe to work.
> > > I had some issue when adding more than one line because it was trying to
> > > copy everything to the same p4 folder.
> >
> > Ah, yeah... Sorry, that is a bit of a shortcoming of my changes.  The
> > reason I made this change is because putting the given SRC_URI's
> > repository directly into a directory like ${WORKDIR}/p4/ is how some of
> > the other fetchers operate and it makes the automatic tasks just work
> > correctly.  This is how the git fetcher does things so I tried to mimmic
> > that but maybe I missed some nuance in how the git fetcher deals with
> > more than one SRC_URI line being a git repo.  I hadn't realized that
> > there was users of the p4 fetcher who did things like you are doing,
> > sorry.
> >
> > I think it should be reasonably easy to make a patch to the p4 fetcher
> > which will put each SRC_URI line which uses the p4 fetcher into its own
> > directory within the ${WORKDIR}/p4/ directory such that it supports your
> > style of fetching.  This then will likely require your recipe to know
> > that this is the directory structure and properly change into
> > directories as needed, but it sounds like you already comprehend that.
> > And for users of the p4 fetcher who only use one SRC_URI line, I believe
> > they would just need to set the S variable correctly to the directory
> > name within ${WORKDIR}/p4/ to get the current operation.
> 
> I have been looking at this and it wasn't straight forward for me... I
> don't have much experience with this. I will continue trying but I might
> need some help.
> It's not only not being able to fetch more than one SRC_URI. We are also
> loosing the flexibility of fetching different changelists for each line.
> I'm not sure if this can be solved in this new fetcher whilst keeping git
> and p4 "similar" as git doesn't work this way afaik.

It sounds like you really just want to revert to how the p4 fetcher used
to work.  My changes were because I wanted the p4 fetcher to work more
like the git and svn fetchers.  But that doesn't sound like what you
want.

My p4 fetcher changes which are impacting you on poky's pyro are these:

35081f55185a8a9804c21b16cd4600ba70646a10 bitbake: bitbake-user-manual: Addeds support for the Perforce Fetcher
40f3099e8ec5c6f3a8b7f3d0e90f1c65c388ee77 base.bbclass: p4 fetcher supports srcrev
26447a0d0bf5ceaac382a78c1dbd00807dbab706 bitbake: fetch2/perforce: Rework to support SRCREV and P4CONFIG

But that last one will run into conflicts if you just try to revert it
directly due to having other commits since then which git can't easily
figure out, but if you revert all of these commits too, then you may
have to redo some of the changes they made which are related to general
things and are not p4 specific.  Commits you'll want to look at are:

b16192c93834d0a6530169557aa34122e1417bcf bitbake: fetch2: don't use deprecated bb.data APIs
6c611d697f9c03867c938cba1b481f38eebed8bf bitbake: fetch2: Rename "setup_revisons" to "setup_revisions"
38438b6cf42fb7ad45b9a901f57913af7e7591a3 bitbake: fetch2: obey BB_ALLOWED_NETWORKS when checking network access
1fce7ecbbb004a5ad82da3eef79cfd52b276708d bitbake: bitbake: remove True option to getVar calls
ab09541d5517da9b1a23923ea8f5c26ddf745084 bitbake: fetch2: preserve current working directory
26447a0d0bf5ceaac382a78c1dbd00807dbab706 bitbake: fetch2/perforce: Rework to support SRCREV and P4CONFIG

I haven't actually reviewed any of these commits after the "preserve
current working directory" one as I'm still doing my day-to-day work on
the morty branch of oe-core and bitbake 1.32.  But possibly one of these
commits are unintentionally impacting your ability to specify SRCREV and
that's worth checking into.

Thanks,
Andrew