> Why you changed `printSymbol: aClass name`
> for `nextPutAll: aClass name`?

Because there is no #printSymbol: anywhere in Pharo 7 or Pharo 8> ,

so I had to *guess* what it does, and the only reason I could see for

using #printSymbol: instead of #print: is that class names are symbols

and #print: will write a hash in front, and it seemed plausible that

you would not want that hash.�� Accordingly I chose a well known, even

standard, method that writes a class name with no extra decoration.

> Now it outputs things incorrectly. You
> should have changed it for `print: aClass name asSymbol`

How is one supposed to *tell* what "correct" output is?

If I wanted to write that code, I would be kind to my readers

and make it *obvious* by doing

���� stream nextPut; $#; nextPutAll: aClass name.

I certainly would not go out of my way to confuse people by

sending #asSymbol to something they already know is a Symbol.

For what it's worth, I've attached my implementation.

On Wed, 11 Sep 2019 at 21:46, Herby Voj��k <herby@mailbox.sk> wrote:

On 11. 9. 2019 3:23, Richard O'Keefe wrote:
> It is good that you have a coherent idea of how << can work.

After the changes sent by Sven, Pharo 8 seems to have the exactly same
idea. IMNSHO.

> The question I was addressing is how << *DOES* work in Pharo.
> Having simple things working, if they are kept consistent,
> is indeed good.�� The problem is that in Pharo 7 they are NOT
> consistent, and << does NOT work consistently, because there
> are occasions when << does ^something nextPutAll: something else
> and nextPutAll: returns something else.

Bug. '^' should not have been there.

> I am grateful for the link to the changes to Pharo 8.
> That's a massive simplification and improvement.
> There is some way to go yet.

Actually, from my PoV, just the fix I posted. The rest seems correct.

> The very first thing that should be done is to write down
> what the *semantics* of #<< is supposed to be, and then to
> ensure it is implemented that way. >
> Let's look at the chunk example.
>
> exportTraitDefinitionOf: aClass on: aStream
>�� "Chunk format."
>�� aStream
>�� nextPutAll: 'Trait named: '; nextPutAll: aClass name; cr;

The devil is in the details. Why you changed `printSymbol: aClass name`
for `nextPutAll: aClass name`? Now it outputs things incorrectly. You
should have changed it for `print: aClass name asSymbol`

And this is precisely that distinction between low-level nextPutAll: and
high-level write: / << / print:.

>�� tab; nextPutAll: 'package: '; print: aClass category; nextPutAll:
> '!!'; cr.
>�� aClass comment isEmpty ifFalse: [
>�� aStream
>�� nextPutAll: '!!'; print: aClass; nextPutAll: 'commentStamp!!'; cr;
>�� nextPutAll: aClass category withDelimiter: $!; nextPutAll: '!!'; cr].
>�� aStream cr.
>
> With the exception of #nextPutAll:withDelimiter:, this is completely
> standard and portable.�� If I am willing to do something nonstandard,
>
>�� aStream format: 'Trait named: {s}{n}{t}package: {p}{n}'
>�� with: aClass name with: aClass category.
>�� aClass comment isEmpty ifFalse: [
>�� aStream format: '!!{p} commentStamp!!{n}{D$!}!!{n}'
>�� with: aClass with: aClass comment].
>�� aClass format: '{n}.
>
> #write: really does not seem to be any improvement over #nextPutAll:.

Will post.

> For what it's worth, GNU Smalltalk also has #<<, which it defines
> to be equivalent to #displayOn:, if memory serves me correctly.
> It has never been clear to me what the use case for #write: is.
> In Pharo 7, for streams it is the same as #putOn: except for the
> result.�� Neither of them is in any interesting way "higher
> level" than #nextPut: or #nextPutAll:, merely less informative.