Why doesn't Pharo 7 Iceberg write changes to the git filesystem as you go?
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works. When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen? When you branch in Pharo, a command line git status does show that change - so some things clearly are being mirrored, just not code (Which Iâm guess happens briefly when you click commit?). Iâm curious now to understand the tradeoffs. Tim p.s. it is very nice for small private projects, to use a git client on your phone - edit a method or two on the train, commit your changes and then see your CI build the results and deploy a new website by the time you get off⦠yes its not the rich smalltalk environment for bigger changes - but tiny stuff, its quite nice to fallback on the traditional way.
Partial answer, but you saw this, right ? https://github.com/pharo-vcs/iceberg/wiki/Iceberg-glossary
On 22 May 2018, at 17:23, Tim Mackinnon <tim@testit.works> wrote:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
When you branch in Pharo, a command line git status does show that change - so some things clearly are being mirrored, just not code (Which Iâm guess happens briefly when you click commit?).
Iâm curious now to understand the tradeoffs.
Tim
p.s. it is very nice for small private projects, to use a git client on your phone - edit a method or two on the train, commit your changes and then see your CI build the results and deploy a new website by the time you get off⦠yes its not the rich smalltalk environment for bigger changes - but tiny stuff, its quite nice to fallback on the traditional way.
Actually I missed that one - thanks. Would have been a useful one to show everyone when we went through the commit process - itâs very good. Sent from my iPhone Sent from my iPhone
On 22 May 2018, at 17:16, Sven Van Caekenberghe <sven@stfx.eu> wrote:
Partial answer, but you saw this, right ?
https://github.com/pharo-vcs/iceberg/wiki/Iceberg-glossary
On 22 May 2018, at 17:23, Tim Mackinnon <tim@testit.works> wrote:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
When you branch in Pharo, a command line git status does show that change - so some things clearly are being mirrored, just not code (Which Iâm guess happens briefly when you click commit?).
Iâm curious now to understand the tradeoffs.
Tim
p.s. it is very nice for small private projects, to use a git client on your phone - edit a method or two on the train, commit your changes and then see your CI build the results and deploy a new website by the time you get off⦠yes its not the rich smalltalk environment for bigger changes - but tiny stuff, its quite nice to fallback on the traditional way.
I agree with you that it is weird, but as I understand it it is because the process of "Saving" a package and "Committing" it is merged into a single action, so there is no way to "save" the package other than committing its changes to the repo. This has the drawback you mentioned, but I guess this enables you to simultaneously save the same package to more than one repository. Maybe a diagram could make the whole picture more clear. Regards, Esteban A. Maringolo 2018-05-22 12:23 GMT-03:00 Tim Mackinnon <tim@testit.works>:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
When you branch in Pharo, a command line git status does show that change - so some things clearly are being mirrored, just not code (Which Iâm guess happens briefly when you click commit?).
Iâm curious now to understand the tradeoffs.
Tim
p.s. it is very nice for small private projects, to use a git client on your phone - edit a method or two on the train, commit your changes and then see your CI build the results and deploy a new website by the time you get off⦠yes its not the rich smalltalk environment for bigger changes - but tiny stuff, its quite nice to fallback on the traditional way.
Esteban A. Maringolo wrote
there is no way to "save" the package other than committing its changes to the repo.
In the past, I sorely missed a feature like this. When Iceberg (or I!) was confused about the state of the local repo/changes/whatever, I wished I had a way to write to disk and then verify that it did "the right thing" and manually fix with "normal" git tools if needed. Maybe Iceberg has improved enough to make this unneeded⦠----- Cheers, Sean -- Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
On 22 May 2018 at 23:23, Tim Mackinnon <tim@testit.works> wrote:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
I guess the conceptual model they have might be of Pharo as a text editor directly changing the files. Its interesting to consider how that would operate. Changes are immediately reflected in Epcia files, and used to be immediately written to ".changes" file, so it seems possible. More interesting is what to do if the text file is edited outside Pharo. If Pharo could observe this and then recompile the new file, we might suddenly have a workable "external editor" workflow, lack of which is a common complaint by some. cheers -ben
When you branch in Pharo, a command line git status does show that change - so some things clearly are being mirrored, just not code (Which Iâm guess happens briefly when you click commit?).
Iâm curious now to understand the tradeoffs.
Tim
p.s. it is very nice for small private projects, to use a git client on your phone - edit a method or two on the train, commit your changes and then see your CI build the results and deploy a new website by the time you get off⦠yes its not the rich smalltalk environment for bigger changes - but tiny stuff, its quite nice to fallback on the traditional way.
On Wed, May 23, 2018 at 6:42 AM, Ben Coman <btc@openinworld.com> wrote:
On 22 May 2018 at 23:23, Tim Mackinnon <tim@testit.works> wrote:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
I guess the conceptual model they have might be of Pharo as a text editor directly changing the files. Its interesting to consider how that would operate. Changes are immediately reflected in Epcia files, and used to be immediately written to ".changes" file, so it seems possible.
This is indeed valuable when things go wrong.
More interesting is what to do if the text file is edited outside Pharo. If Pharo could observe this and then recompile the new file, we might suddenly have a workable "external editor" workflow, lack of which is a common complaint by some.
I don't believe that it should be Pharo's concern to waste effort on supporting external editors -- it is throwing so much value of Pharo. That being said, until git conflicts can be resolved inside Pharo, it is necessary to be able to edit the files by hand. The ability to run the hand-resolved code without committing it (to verify it still works/compiles) seems like a worthwhile feature. However if you then edit some methods - and then look in the file system -
git status doesnât show these? This in retrospect possibly feels weird - or does it?
With Iceberg you have two separate working copies. It is not weird from git perspective, however it is a rather unique setup -- even seasoned git users might say... "You can do that? :-o" :) (And yes, it adds some extra cost, but nothing is perfect.) Peter
2018-05-23 15:28 GMT+02:00 Peter Uhnák <i.uhnak@gmail.com>:
On Wed, May 23, 2018 at 6:42 AM, Ben Coman <btc@openinworld.com> wrote:
On 22 May 2018 at 23:23, Tim Mackinnon <tim@testit.works> wrote:
Hi - when trying out the new Iceberg with a bunch of developers and explaining the challenges of integrating git and files into a smalltalk realm of the image - there was a lot of interest in how this works.
When you clone - you obviously see a series of files (in Tonel - nice) that are then brought into your image. If you edit a file like Readme.md (using a markdown editor) you will notice that git status will show you that this file has changed. However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it? Iâm not sure anymore - and was wondering if there was a specific reason behind not mirroring code changes back to the file system as they happen?
I guess the conceptual model they have might be of Pharo as a text editor directly changing the files. Its interesting to consider how that would operate. Changes are immediately reflected in Epcia files, and used to be immediately written to ".changes" file, so it seems possible.
This is indeed valuable when things go wrong.
At a point, I was considering using git as the changes file (every method save and compile in Pharo -> a commit in git) Didn't try to see if git would handle the load.
More interesting is what to do if the text file is edited outside Pharo. If Pharo could observe this and then recompile the new file, we might suddenly have a workable "external editor" workflow, lack of which is a common complaint by some.
I don't believe that it should be Pharo's concern to waste effort on supporting external editors -- it is throwing so much value of Pharo.
That being said, until git conflicts can be resolved inside Pharo, it is necessary to be able to edit the files by hand. The ability to run the hand-resolved code without committing it (to verify it still works/compiles) seems like a worthwhile feature.
Good point.
However if you then edit some methods - and then look in the file system - git status doesnât show these? This in retrospect possibly feels weird - or does it?
With Iceberg you have two separate working copies. It is not weird from git perspective, however it is a rather unique setup -- even seasoned git users might say... "You can do that? :-o" :) (And yes, it adds some extra cost, but nothing is perfect.)
I considered eliminating one of the working copy when in Pharo, by cloning repositories in RAW mode (without working tree). What you wrote (the git fast-import layer) allows you to do that. But, to be able to pull, you need to be able to be able to do a merge (and resolve conflicts)... in Pharo memory. Thierry
Peter
participants (7)
-
Ben Coman -
Esteban A. Maringolo -
Peter Uhnák -
Sean P. DeNigris -
Sven Van Caekenberghe -
Thierry Goubier -
Tim Mackinnon