hpi-swa / squot Goto Github PK

Working with diffs instead of whole snapshots should speed up regular operations such as saving new versions.

Author: Fabio Niephaus

Message:
We talked about that

The merge base should be taken into account, so Squot does not propose to revert the current changes that do not come from the merged version.

Author: Fabio Niephaus

Message:
It would be great to be able to provide feedback straight from the UI. Maybe you can find someone to add this feature for you, while you are busy implementing more important features... ;)

Author: Fabio Niephaus

Message:
Instead of fetching from all remotes, the user could be ask from which remote to fetch when clicking the Fetch button. Everything could be selected in the list by default, so one also needs to click ok to fetch from all

Untrack files also need to be merged into a branch.
Currently, "untracked" changes on a feature branch get lost when the branch is merged.

Author: Fabio Niephaus

Message:
When '.git/refs/remotes' is missing, I cannot open the UI, because of an FileDoesNotExist exception.

It should be possible to "ignore" a tracked object (like gitless allows). This could be toggled on the object itself, for example in the commit dialog. It is also desirable to ignore only part of a tracked object (e. g., a single method of a package artifact).

Since we do not use a traditional git working copy, there is no reflog for HEAD. To not lose track of what is being checked out or committed in in-image working copies, a reflog should be created for each working copy in the image.

How to name this reflog? Should there also exist a symbolic reference for this working copy? Look up how Gitless stores its heads (one per branch).

Author: Fabio Niephaus

Message:
Test

Author: @krono

Message:
Put a selector to #enabled: for each button, for example for "Commit", so that it is not enabled when there's no proper selection

And more importantly, removing the SquotWorkingCopy does not make the image shrink again. Investigate what holds onto blobs and packs and implement something to prevent keeping large byte arrays.

Author: Fabio Niephaus

Message:
There still is a halt in the actionFetch code path.

Author: Fabio Niephaus

Message:
Start a new project, add first commit, then push. This results in a MNU UndefinedObject>>parents in GitHistoryWalker>>visitCommit:. Apparently, the commit is nil...

During history walking, a lot of time is spent in the garbage collector. The code should be investigated to find places where objects can be reused or where objects are created unnecessarily. Also consider more aggressive caching.

Author: @fniephaus

Message:
I am getting a lot of DirectoryDoesNotExist and Unable to open ... errors. Interestingly, it seems to be working for a short period of time. Once one of these errors occur, all future file system access operations fail. Reopening the image seems to "fix" this, but I'm getting an error again, as soon as I try to commit something :(

Probably v0.1.0 or something like that, and then we should follow http://semver.org/.

Author: Fabio Niephaus

Message:
Wait for it

The resulting packfile is corrupt.

Workaround: do not pack the repository with FileSystem-Git. Use the regular git client instead.

GitHub terminates the connection if nothing is supplied quickly enough. Pushing the first 361 commits of this repository did not work with FileSystem-Git. I had to use the non-Smalltalk git client instead.

Workaround: reset packs inst var to nil and restart the method looking into the packs.

...and cannot be resolved subsequently. Investigate what goes wrong.

Depends on #3

Following up on this discussion: how should the tracking work? Shall it track categories, packages, both, ...?

(NeedGitCredentials for: url) signal does not do anything in my image.
Where do I have to add my credentials?

There should be a button in the UI to temporary clone the Squot repo and load the latest master version from it, so the SWE students can easily update Squot&Co.

The other path would be to package Squot&Co. as Monticello versions and provide those via the update stream. This would need at least a simple GitCommit --> MCVersion(s) conversion.

Should be labelled "reset to this commit".

Do not touch the object memory, only do loadedHistorian version: theSelectedSquitVersion, to move the ref. This can be used to manually fast-forward the master to a remote tracking branch (origin/master), until a proper merge operation (with fast-forward capability) is implemented.

Author: Fabio Niephaus

Message:
Add menu items for merging to branch and commit list

When adding a new project, the tool does not display that there is a master branch. Not sure what needs to be done here.

Author: Fabio Niephaus

Message:
This would help reading the list

I did push successfully twice to GitHub with FileSystem-Git today, but for some sets of commits, GitHub simply closes the connection after receiving a part of the pack content. For the same set of commits, both a local test server and Bitbucket accepted the push without issues.

Have to keep an eye on this to find out what is the cause.

It returns only the directory, but not the refs inside.

Currently, done is sent immediately, hence a remote will always send the whole repository back as a packfile. Proper negotiation should be implemented to reduce traffic (and fetch time) as described in pack-protocol and http-protocol

Author: Fabio Niephaus

Message:
Can I haz pull button

When I click fetch, I get an "Unknown stream code 32" error in readSideBandResponseIfData:ifMessage:ifError:ifZero:.
Also, I get an ConnectionClosed error when trying to submit feedback.
However, WebClient httpGet: 'https://google.com' works as intended. Not sure what is going on...

In this case, the iteration should stop at the bottom commit, but it does not, because only remotes/origin/master was designated as a sink commit. In effect, too many objects are added to the pack.

A tree view would work well in the commit dialog.
It could be used to display per method changes per class. Maybe, you can also find a way to display all changes for a class or even one big diff for everything.

While it is mandatory to allow the set of pack files to change over time, rediscovering them over and over during an operation such as walking the history is unnecessary. There should be a way to delimit an operation and cache the pack file references and the pack index files' content during that operation.

branch.name.remote is the remote name and branch.name.merge is the reference at the remote

Create a new class and add it to some new category. Then start a new project in the GitBrowser. Currently, you cannot add the category, because it does not appear in the package list. In my current image, I have replaced this list with the same list the Browser displays. However, when I add the newly created category, a new commit is empty. Have tried different things, can't get it working

Inspecting the VM process with Process Explorer revealed that the file handles for pack files stay open, even though the corresponding FSHandle instances are already garbage collected or closed, as far as the object memory is concerned.

@fniephaus can you please check if the same is true for Mac OS? Just work a while in your image and check the process after some git operations.

Snippets to find (officially) open files:

FSStream allSubInstances select: [:ea | (ea instVarNamed: 'handle') reference filesystem = FileSystem disk]

FSFileHandle allInstances select: [:ea | ea isOpen]

The tools to inspect the actually open file descriptors or handles vary by OS, of course.

Currently, the default behavior for Object for taking snapshots is to make a #copy. This breaks the meta object protocol because #allInstances will include the objects contained in snapshots.

Author: Fabio Niephaus

Message:
Since Squot does not support git+ssh at the moment, it would be good to somehow ensure a user can only enter HTTP(S) URLs.

The commit dialog shows up empty and the diff content is empty as well.

The MCPatch is empty. It should include additions for all definitions in the package.