[squeak-dev] Unambiguously differentiating official and local builds [Was [Vm-dev] Moving the Cog subversion repository to githup at 2016-6-16 7am UTC]

Sat Jun 18 12:51:21 UTC 2016

-- 

On Sat, Jun 18, 2016 at 11:06 AM Ben Coman <btc at openinworld.com> wrote:

>
> On Sat, Jun 18, 2016 at 11:28 AM, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
> >
> > Hi Ben,
> >
> > On Fri, Jun 17, 2016 at 8:27 AM, Ben Coman <btc at openinworld.com> wrote:
> >>
> >>
> >> On Fri, Jun 17, 2016 at 3:40 PM, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
> >> >
> >> >> On Jun 17, 2016, at 12:22 AM, Tim Felgentreff <
> timfelgentreff at gmail.com> wrote:
> >> >>
> >> >>> On 16 June 2016 at 22:07, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
> >> >>> Hi All,
> >> >>>
> >> >>>     so after fixing "git remote get-url origin" to fail over to
> "git remote
> >> >>> show origin | filter and munge" the culture shock of "git commit
> -a" (git
> >> >>> commit does nothing ?!?!?) I have a VM that outputs a reasonable
> version
> >> >>> info:
> >> >>>
> >> >>>
> /Users/eliot/oscogvm/build.macos32x86/squeak.cog.spur/CocoaFast.app/Contents/MacOS/Squeak
> >> >>> 5.0 5.0.201606161953 Mac OS X built on Jun 16 2016 12:56:52 PDT
> Compiler:
> >> >>> 4.2.1 Compatible Apple LLVM 6.0 (clang-600.0.57) [Production Spur
> VM]
> >> >>> CoInterpreter VMMaker.oscog-eem.1886 uuid:
> >> >>> d413db9f-37cc-4c5d-bfc6-87b11203ee96 Jun 16 2016
> >> >>> StackToRegisterMappingCogit VMMaker.oscog-eem.1886 uuid:
> >> >>> d413db9f-37cc-4c5d-bfc6-87b11203ee96 Jun 16 2016
> >> >>> VM: r201606161953 http://github.com/OpenSmalltalk/vm $ Date: Thu
> Jun 16
> >> >>> 12:53:33 2016 -0700 $
> >> >>> Plugins: r201606161953 http://github.com/OpenSmalltalk/vm $
> >> >>>
> >> >>> Which begs the question how do I differentiate this from something
> built
> >> >>> officially via Travis?  Arguably the URL is wrong, and should only
> say
> >> >>> "http://github.com/OpenSmalltalk/vm" for travis builds, and
> perhaps should
> >> >>> just include my local hostname and current directory when I make
> any kind of
> >> >>> local modification.  So the above would read
> >> >>>
> >> >>> ...
> >> >>> VM: r201606161953 McStalker:?users/eliot/oscogvm $ Date: Thu Jun 16
> 12:53:33
> >> >>> 2016 -0700 $
> >> >>> Plugins: r201606161953 McStalker:?users/eliot/oscogvm $
> >> >>>
> >> >>> Alternatively we could add another field, or modify one of the
> existing
> >> >>> fields to say "I'm official" however one would do that.  I don't
> know how, I
> >> >>> just know we need this.  I shouldn't be able to pollute the VM pool
> by
> >> >>> putting some VM on some site somewhere that i just happened to
> build after
> >> >>> several sherries and some cannabis brownies that looks to all
> intents and
> >> >>> purposes just like a VM built by our official Travis slaves.  Hic.
> Chillin'
> >>
> >> I just discovered git-describe, which seems like it could be useful...
> >>
> http://alblue.bandlem.com/2010/11/automatically-tagging-builds-with-git.html
> >>
> >> So if Travis created  "r201606161953" as an *official* tag for
> >> successful builds like this...
> >> https://github.com/travis-ci/travis-ci/issues/1476
> >>
> >> then `git describe` would produce "r201606161953" for that build, and
> >> after a couple of commits in my personal repo would produce
> >> "r201606161953-2-g169d02a".    The "-2-g169d02a" would distinguish
> >> non-official builds.
> >>
> >> In addition, I can now copy-paste a VM's output revision string
> >> to directly do "git checkout r201606161953"
> >> instead of "git checkout master@{2016-06-16 19:53} which I read is
> >> only viable for 90 days anyway, and has some complexity between
> >> whether the given date is author commit date or merge date.
> >>
> >> But after doing "git checkout r201606161953" in my personal repo
> >>    git describe
> >>         ==>    r201606161953  is indistinguishable from the Travis build
> >> but...
> >>    git describe --long
> >>        ==>  r201606161953-0-a264e03b  is distinguishable.
> >>
> >> In addition,  if I edit some files and rebuild before committing I
> >> want to distinguish this from when I build a fresh check out , which
> >> can be done with...
> >>    git describe --long --dirty    ==>   r201606161953-0-a264e03b-dirty
> >>
> >> So that last would be used to version personal builds,
> >> while Travis would use "git describe" without any flags.
> >>    ==>   r201606161953
> >
> >
> > Sounds really good, but
> >
> > McStalker.oscogvm$ uname -a
> > Darwin McStalker 13.4.0 Darwin Kernel Version 13.4.0: Mon Jan 11
> 18:17:34 PST 2016; root:xnu-2422.115.15~1/RELEASE_X86_64 x86_64
> > McStalker.oscogvm$ git --version
> > git version 1.9.5 (Apple Git-50.3)
> > McStalker.oscogvm$ git describe
> > fatal: No names found, cannot describe anything.
> > McStalker.oscogvm$ git describe --long
> > fatal: No names found, cannot describe anything.
> > McStalker.oscogvm$
>
> That confused me also for a moment.  Its only because there are not
> yet any tags.
> Try this...
> $ git tag -a mytag -m "my message"
> $ git describe --long --dirty
> $ touch x
> $ git add x
> $ git commit -m "blah de blah"
> $ git describe --long --dirty
> $ echo blah > x
> $ git describe --long --dirty
>
> Plus these tags can be applied retroactively without affecting history...
> https://git-scm.com/book/en/v2/Git-Basics-Tagging
>
> cheers -ben
>

It might be a good idea to tag the last (or first?) commit after the
migration. But how do we do tagging? Have we decided on anything yet (e.g.
[1])? Shall we start with v1.0.0 or shall v1.0.0 be the very first stable
release of the OpenSmalltalk VM? Or would `201606171704` be a tag? We
certainly shouldn't confuse people with these numbers and the version
number of the VM.

Cheers,
Fabio

[1] http://semver.org/

>
> >
> >> >> how secure does this need to be? One way to differentiate the
> official
> >> >> VMs is to sign them directly on Travis (which we'll want to do
> anyway,
> >> >> just didn't get to it, yet).
> >> >>
> >> >> Another option is to just change the URL replacement code to do
> >> >> something else when not running on Travis --- like adding your
> >> >> hostname and path instead --- but this could be fairly easily messed
> >> >> with.
> >> >>
> >> >> Not sure how much malicious intent we want to prevent.
> >>
> >> Later on we should have Travis signing its build artefacts, but for
> >> now keep it simple.
> >
> >
> > The Mac builds already sign provided a certificate is installed and an
> environment variable set to point to it.  See SIGNING_IDENTITY in
> build.macos*/common/Makefile.app
> >
> >>
> >>
> >> >
> >> > None.  I don't think there's malicious intent at all.  I do think we
> should differentiate between "personal" and Travis builds.  It's more for
> my own information, so u don't get confused, than to prevent
> maliciousness.  So do the simplest thing that could possibly work TSTTCPW.
> I like username,host name,path as in an scp, eg eliot at McStalker:oscogvm
> (path relative to ~eliot).
> >>
> >> I think `git branch` is as important as `path`.
> >> Username could come from `git config user.name | sed 's/ //g'
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20160618/91ac14f2/attachment-0001.htm