Le 04/02/2016 14:32, Nicolas Cellier a écrit :
2016-02-04 10:26 GMT+01:00 Thierry Goubier <thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>>:
Le 04/02/2016 10:04, Nicolas Cellier a écrit :
[... cut ]
Or is it the fact that some .mcz could be missing? Consider this is a feature, MC tools are robust to missing .mcz (it's just that you'll have to redo the merge if you lost a common ancestor).
What? You really consider that a feature?
Sure. Consider it's somehow like a git squash...
I consider history rewriting a very bad idea... But, if we are at that, I just consider that the removed versions do not exist anymore, and are not visible (should not be visible) in the history, and that you have perfectly predictable correct behavior, not can't do this because package version X is missing.
Or is it the fact that some repository might contain only a slice of history? This is another feature... You can view all the versions in a collection of repositories without needing to replicate.
Understandable in theory. Unworkable over time and change (repositories disappear and die, and this stops working)
But that's the same with over vcs, repositories disappear and die (google code etc...). Knowing that you can take usual defensive decisions (replicate what you depend on and not trust too much).
I mean that VCS recreate a working history and log when you are in such a situation. MC just let you believe you have all and fail when you attain the holes.
You can replicate if you want but it's not mandatory and completely orthogonal. So yes, this information - the list or repositories you want to consider - has to be stored separetely and this can sound quite unconventional. But IMO, it's an important feature: it gives much resilience for a very low investment. And that also mean that you can hardly break things (have unconsistent history).
Maybe when you say broken, you mean not 100% git compatible?
No, what I say is true mcz inconsistent history (missing versions making merges very unreliable, basically).
Loosing a .mcz on which you branched is a bit more than unfortunate, it's a lack of understanding of the tool...
Not true. MC let you believe it will cope and it doesn't, which is not exactly the same.
I say give git squash and rebase to unexperimented/uninformed people and that will be far worse.
Agreed.
I describe a while ago a case where, thanks to mcz features, I couldn't merge a small change done to Roassal without generating a ton of conflicts. I moved the three needed mcz(s) to git (the change ancestor, the change, and the current head), did git merge and had no conflicts.
Do you mean that git succeeded in merging text because different lines were changed in different branches?
No, in fact it was a clean merge because the changes in the branch were on methods untouched in the main... git recreated a proper history and could merge, whereas MC with the mcz history generated conflicts without any reason.
I agree that it can solve some basic refactoring (renaming...) But couldn't it lead to syntactically incorrect code? Traditional Smalltalk IDE are not well equipped for handling this...
Yes, this is partly why we're looking at more integration (merge tool) and we may have to do more (a smalltalk syntax checker before commit could be nice; CI does wonders on that).
The biggest grief I have with MC is more about package delimitation, when code is moved from 1 package to another. For example, this effectively prevent merging very distant branches like Squeak and Pharo (but we do not want to merge, just cherry pick in fact).
Ok. Merge with multiple packages, right?
If you consider those features, then I disagree.
I'd really like to improve MC and get a better integration. Considering those as features just make me think that Eliot may well be right in believing we'll end up throwing away MC completely.
As long as it's as efficient, and well integrated in IDE, but again, it's against the small steps approach... For me small steps would be to replace mcz name indexing by UUID indexing and just keep name as a façade.
I'd agree with that, as long as you don't ask too much of compatibilities with some of MC features (:)).
I'm quite sure most work is at server side. How difficult would it really be?
Why server side? I don't think it is.
No one will consider this maybe, because it's against the stream ;)
No, in fact we're on it. Git integration is showing us where MC is weak, and we're trying to fit it with MC.
Coping with MC idiosyncrasies is hard as it is, and at least a complete replacement is being considered for Pharo. In the meantime, Cuis has completely given up managing packages in Smalltalk: all is done in git, externally.
But Cuis vcs come from dinosaure era (change sets) so anything else is already a progress. Doing that in Squeak would feel like a loss of tool/functionality/integration.
I agree.
Why didn't Pharo took the same path?
Than Cuis? Thierry
Thierry
2016-02-04 7:53 GMT+01:00 Thierry Goubier <thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com> <mailto:thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>>>:
Le 03/02/2016 23:58, Dale Henrichs a écrit :
On 02/03/2016 02:34 PM, Thierry Goubier wrote:
Le 03/02/2016 22:51, Eliot Miranda a écrit :
On Wed, Feb 3, 2016 at 12:54 AM, Thierry Goubier <thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com> <mailto:thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>> <mailto:thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>
<mailto:thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>>>> wrote:
Hi Eliot,
Le 02/02/2016 21:54, Eliot Miranda a écrit : ....
No it's /not/ the end of the story. The essential part of the story is how Monticello remains compatible and interoperable between dialects. I haven't seen you account for how you maintain that compatibility. As far as I can tell, you propose replacing the Monticello metadata with that from git. How do I, as a Squeak user with Monticello, ever get to look at your package again? As I understand it, moving the metadata from Monticello commit time to git means that the metadata is in a format that git determines, not Monticello.
Yes. See below why.
So I don't understand how on the one hand you can say "The Monticello metadata in a git repository is redundant and leads to unnecessary commit conflicts -- end of story ....", which implies you want to eliminate the Monticello metadata, and on the other hand you say you're keeping the Monticello metadata. I'm hopelessly confused. How does the Monticello metadata get reconstituted if it's been thrown away?
What happens to the metadata in the following workflow?
load package P from Monticello repository R into an image change P, commit via git to local git repository G load P from G into an image store P to R via Monticello
It's not a scenario I've specifically worked on, but all the tech is implemented / implementable to do that perfectly.
The only thing that is problematic there is that the only safe history is the one generated from git... there are so many MC packages with broken history that, on mcz packages, you have to admit that it's not safe to base things on their history.
I'm sorry but I don't accept that. In the Squeak trunk we have history in our mczs that is correct. Certainly in VMMaker.oscog I have history that goes back a long time. If bugs have broken history then efforts should be made to repair that history. But you can't just write off Monticello history like that.
I don't. You presuppose.
I write tools that work with Monticello repositories, not just yours. I have to do with what is given to me. On a general level, as a mcz user, I'll just have to consider that you are as susceptible to be trusted as with any other mcz producer. This means not much...
... And this is the reason why I am inclined to favor option 3, which records the package version history as it existed at the point it was copied into a git repo. When copied back out from the git universe, create a version history that starts with the original version history and generates a history of the package in git ....
Which is not very difficult to do given how GitFileTree is implemented. And I agree this may well be the way to go. But ...
Correct or not, the Monticello version history should be preserved ....
I wonder about that. The property of the Monticello version history is that it has value when you can access the versions listed in it. If you mix repositories like that, unless you maintain a link to the previous mcz repository, pre-git versions can't be accessed. So, most of the time, what we do with a project moved under git is to clone the previous repository, not take just the current head.
(you'll notice, by the way, that vcs usually work that way when moving from, say, CVS to git - you move the entire repository, not just the latest version).
In short, the question would really be: 1- should we invest into making that integration of past history a selling point (but I foresee issues down the road; I've only described one so far, and I have seen others) or 2- into making a better "clone" of a package history, timestamps and everything when moving a complete repository to git?
Honestly, I'd consider 1- to be the easiest to implement. 2- there is already some code floating around.
Thierry