[Vm-dev] [OpenSmalltalk/opensmalltalk-vm] 9fd4e3: Use LF instead of CR as image/*.st line ending

[Ben Coman]

On Fri, Apr 28, 2017 at 1:47 PM, Eliot Miranda <eliot.miranda at gmail.com>
wrote:

>
> Ben,
>
> On Apr 27, 2017, at 7:06 PM, Ben Coman <btc at openinworld.com> wrote:
>
> On Fri, Apr 28, 2017 at 6:16 AM, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
>
>>
>> On Thu, Apr 27, 2017 at 2:38 PM, Nicolas Cellier <
>> nicolas.cellier.aka.nice at gmail.com> wrote:
>>
>>>
>>> Hi Eliot,
>>> it's not about my favourite editor, it's about the universally available
>>> one:
>>>
>>> https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/57c
>>> 9ef837f7b0914351c4836734d1df3c880c288
>>>
>>> The LF are displayed correctly in Squeak/Pharo and we can use nextLine
>>> in replacement of upTo: Character cr, so I fail to see the problem of LF.
>>> Could you explain?
>>>
>>
>> It's ok.  I give up.  This is another example of the tail wagging the
>> dog, which gets on my nerves.  But getting them to fix their interface is
>> way more difficult than us accommodating their limitations so go ahead.
>>
>
> Its not github!  When looking into Sophie's issue, opening those ST
> scripts were unreadable
> when opened in `vi` .  It was a pain that I had to break context to hunt
> for a GUI text app to read them
>
>
I'd advocate going to LF EOL, but if its really important, a bit of
research dug up how git can work with CR-line-ending
Here is a recipe to demonstrate the effect.

First restore one commit back from "Use LF instead of CR as image/*.st line
ending"
$ git status    # && git stash  # if required
$ git checkout 9fd4e371~1
HEAD detached at 11d990e
nothing to commit, working directory clean
$ git diff
no output, okay.

Make a change in directory opensmalltalk-vm/image/
$ vi StartReader.st
         insert a line "comment 1"
$ git diff
@@ -1 +1 @@
-Transcript clear.^MSmalltalk snapshot: true andQuit: true.^M[[Processor
activeProcess bindToThreadId: 2]^M     on: Error^M     do: [:ex|].^M
StdioListener new run] forkAt: Processor activePriority + 1
+"comment 1"^MTranscript clear.^MSmalltalk snapshot: true andQuit:
true.^M[[Processor activeProcess bindToThreadId: 2]^M        on: Error^M
  do: [:ex|].^M StdioListener new run] forkAt: Processor activePriority +
1^M

yuck!

Add a git filter for ST files...
$ vi opensmalltalk-vm/.git/config
    append...
        [filter "crLineEnd"]
             clean = tr '\\r' '\\n'
             smudge = tr '\\n' '\\r'
$ vi opensmalltalk-vm/.gitattributes
    append...
        *.st filter=crLineEnd

Commit file so that EOLs are updated to LF internal to git.
In directory opensmalltalk-vm/image/ ...
$ git add StartReader.st
$ git commit -m "apply filter - test only, throw away"

Now make a second change...
$ vi StartReader.st
        insert line "comment 2"
$ git diff
@@ -1,4 +1,5 @@
 "comment 1"
+"comment 2"
 Transcript clear.
 Smalltalk snapshot: true andQuit: true.
 [[Processor activeProcess bindToThreadId: 2]

Better!
I didn't test it on github, but IIUC it just uses `git diff`


It's a three liner to wrap vi in a script that converts from CR to LF,
> invokes vi and then concerts back.  Is it really that problematic?
>

Its not immediately obvious to me how to do that (maybe sed or tr to a
temporary file that vi then edits?)
So "three lines" is not really a measure of the effort required to work out
how to implement it,
and test that doesn't conflict with normal 'vi' usage. All this breaks from
the context of what I was doing.

Anyhow Tobias's vimrc advice works.

On Fri, Apr 28, 2017 at 8:03 PM, David T. Lewis <lewis at mail.msen.com> wrote:

>
> On Fri, Apr 28, 2017 at 09:39:28AM +0200, Tobias Pape wrote:
> >
> > (in .vimrc)
> >
> > set fileformats=unix,mac,dos
>
> Tobias,
>
> Thank you! I can't believe I have gone so many years without ever knowing
> about this simple editor setting.
>
> Dave


Also thx Tobias.  So many little corners of knowledge.
This works good for me.  Still not so good for the next newcomer that
doesn't know this.

CR-only line endings are sooooooo last century, for machines like:
> Commodore 8-bit machines, Acorn BBC, ZX Spectrum, TRS-80, Apple II family,
> Oberon, the classic Mac OS up to version 9, MIT Lisp Machine and OS-9 [1].
> These dogs died a long time ago ;)
>
> Is there something particularly Smalltalk related I'm not aware of that
> makes CR-only line endings preferable?
>
> [1] https://en.wikipedia.org/wiki/Newline
>
>
On Fri, Apr 28, 2017 at 3:48 PM, Nicolas Cellier <
nicolas.cellier.aka.nice at gmail.com> wrote:
>
>
> Ben,
> There's no need to have three ways to do the same thing, line ending.
>
So Smalltalk has chosen one and stick to it. it was CR.
>

Back when Smalltalk was created, I guess CR was the dominant platform line
ending and so was the choice was "align with the dominant convention".  Now
that the dominant convention has progressed, the question arises whether to
follow to reduce friction with other systems.   The trouble of course being
that there are two competing line endings LF & CRLF.


> Thus you will see a message #cr used everywhere in the code base (either
> sent to Character or to a WriteStream to write an end of line).
>

Ahh. I never made that connection before.


> Also, carriage return is used thru a CR variable in TextConstants pool
> dictionary.
>
> The wisdom was that connections to/from outside world (external streams
> like files/sockets...) should do the conversion in/out the Smalltalk space.
> But this has been inequally applied, especially in Squeak/Pharo where the
> zoo of Stream is not so well engineered (it's a mess!).
>
> Since the boundary was messy, I've taken the responsibility to introduce
> the complexity inside the code base by:
> - changing display of LF and adding hacks to display a single end of line
> in case of CR-LF pair
> - introduce nextLine et a few other methods to deal with the zoo of line
> endings
> I'm still not completely sure that it was a good idea... Maybe
>
If we want to change our conventions to be more fashioned, we must get rid
> of cr message in the code base and replace it with something like eol (end
> of line).
>


> Cuis has changed the CR to LF inside the image.
>
>
I guess Cuis aligned closely with git since thats its main mechanism for
its sources (??).
GNU Smalltalk uses LF (its aligned with Unix).
Dolphin uses CRLF (its aligned with Windows)

Theoretically, a move to the Unicode Line Separator U+2038 might solve all
problems,
   http://unicode.org/standard/reports/tr13/tr13-5.html
but practically it doesn't (yet) seem supported by vim or emacs
And of course, there is this about Standards...   https://xkcd.com/927   .

cheers -ben
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20170501/7412883c/attachment-0001.html>