help wanted: normalising LF on tonel for Pharo project

Hi Ben,

Le 11/04/2018 à 16:37, Ben Coman a écrit :

On 11 April 2018 at 05:05, Esteban Lorenzano <[hidden email] <mailto:[hidden email]>> wrote:

    Hi,

    I’ve been wondering how to better fix the problem of having windows
    and linux/macOS people contributing and the fact that files are
    written in their native system format (crlf windows, lf for the rest
    of the world).



    I digged a bit and I found a couple a link that helped me (after
    trying to understand the doc):
    https://stackoverflow.com/questions/170961/whats-the-best-crlf-carriage-return-line-feed-handling-strategy-with-git
    <https://stackoverflow.com/questions/170961/whats-the-best-crlf-carriage-return-line-feed-handling-strategy-with-git>

    and it seems adding a .gitattributes file with this content:

    # Auto detect text files and perform LF normalization
    *text=auto


I see a few posts around that recommend reading http://adaptivepatchwork.com/2012/03/01/mind-the-end-of-your-line/
which about the above line says... "This is certainly better than requiring everyone to be on the same global setting for core.autocrlf, but it means that you really trust Git to do binary detection properly. In my opinion it is better to explicitly specify your text files that you want normalized."

and https://tinyurl.com/ya9xsprx  says "We had a repo with * text=auto, and Git guessed wrong for an image file that it was a text file, causing it to corrupt it as it replaced CR + LF bytes with LF bytes in the object database."

I'm unsure.  Without it the system is subject to different users' different global settings
and I'd guess that may be a more frequent problem than Git guessing wrong. The latter
can be fixed by a user adding an extra  .gitattributes  entry explicitly specifying the file was binary,
whereas the former seems to introduce a confounding factor.
So probably a good line to have.

    *.sttext merge=union eol=lf

    could fix the problem?
    can someone confirm this?


"eol=lf"   looks appropriate...
https://www.scivision.co/git-line-endings-windows-cygwin-wsl/

Most editors on Windows transparently handle LF line endings.
https://en.wikipedia.org/wiki/Comparison_of_text_editors#Newline_support


" merge=union" I am not familiar with, but I read at... https://git-scm.com/docs/gitattributes
"union = Run 3-way file level merge for text files, but take lines from both versions, instead of leaving conflict markers.
This tends to leave the added lines in the resulting file in random** order and the user should verify the result.
Do not use this if you do not understand the implications."

What are the implications of lines being merged in a random order?


btw, has doing a callback from libgitto a custom merge driver in Pharo been considered?
https://libgit2.github.com/libgit2/#HEAD/group/callback/git_merge_driver_apply_fn

There is a merge driver for parts of the filetree format implemented with Pharo, it could be done on a more general basis if the Tonel format exhibit more conflicts than usual.

But (and this is a big "but"), mixed Pharo / other things repositories with very large files to merge could make things very hard on a smalltalk-implemented merge algorithm.

In most (all?) my professional work, this is the case. I have among my projects a mix FPGA design (verilog + vhdl) + C (drivers, runtime) + Smalltalk, and the smalltalk part is small.

btw2, I found (https://githubengineering.com/move-fast/) interesting...
saying... "Despite being a C library, libgit2 contains many powerful abstractions to accomplish complex tasks that Git simply cannot do. One of these features are indexes that exist solely in memory and allow work-tree related operations to be performed without an actual working directory. [...]  With the in-memory index, libgit2 is capable of merging two trees in a repository without having to check out any of their files on disk."

Yes, I considered that for GitFileTree. The current version uses fast-import and archive (resp. to write and read) and in truth could work on a bare repository, without working tree.

Oh, by the way: it also solves the #lf issue, because you do everything the unix way, even on windows: GitFileTree now never touches a file of the host system.

Thierry

On 11 April 2018 at 05:55, Esteban Lorenzano<[hidden email] <mailto:[hidden email]>>wrote:

    or a .iceberg file?

    Esteban

    ps: yep, we need it… we will have it, why not start now?


Do you mean Iceberg would clone a repo, and from its included  .iceberg  file
a matching  .gitattributes  file would be created?
That seems like double handling.
Why not have the user edit the  .gitattributes  file directly from Iceberg?
Iceberg might provide some appropriate templates.


cheers -ben

Le 12/04/2018 à 03:54, Ben Coman a écrit :

I was thinking that a smalltalk-implemented merge algorithm would only be used for the Smalltal/Tonel code,
not for any other files.  And maybe, when a merge is invoked from Iceberg, the callback to the merge-driver
might present conflicts in a GUI to be resolved, but I guess such would require a threaded-VM.

Two things then.

- What happens if the C developer does a merge in a multi-language project containing tonel files? 

- What is the difference with setting and provides a merge driver in Git, which has the ability to work even without libcgit?