Re: [PATCH 1/3] ob-table: Fix org-sbe's handling of quotes in arguments

[Vladimir Panteleev <thecybershadow@gmail.com>]

On 2018-03-14 15:00, Nicolas Goaziou wrote:
> Vladimir Panteleev <thecybershadow@gmail.com> writes:
>> I wrote about this in the cover letter too. $"foo" and $ "foo" are
>> both the same thing.
> 
> Just to make it clear: I read the cover letter. My confusion doesn't
> come from the fact I may not have read it.

In that case, I need to apologize for not making it sufficiently clear, 
as we're reiterating the same points without making much progress.

> I disagree. You are testing an implementation detail here: the fact that
> "$" is not necessarily a prefix. According to the docstring, it should
> be, so the test should use that, too. What if we rewrite `org-sbe' at
> some point?

I'm sorry if I didn't explain it properly in my previous messages. I'll 
try again.

The fact that $"foo" and $ "foo" mean the same thing is not an 
implementation detail of org-sbe. It is a consequence of Emacs Lisp grammar.

Try evaluating '($"foo") and '($ "foo") in M-:. You will get the same 
result, because after tokenization, their representation is identical. 
The whitespace between $ and "foo" is completely insignificant.

In fact, if you try to evaluate $"foo" by itself, you will get "Trailing 
garbage following expression", because, as I mentioned before, $ and 
"foo" are two distinct tokens.

There is absolutely nothing that anyone could change in org-sbe that 
would make $ "foo" mean something other than $"foo".

So, whether the test case has a space between $ and "a\"b\"c" is as 
relevant as whether it has comments, or uses tabs instead of spaces for 
indentation.

I hope this explanation can put this issue to rest.

> The same goes for the next string. $"foo", or in your case, $"a\"b\"c"
> means nothing in `org-sbe' context. A reference should follow the dollar
> character, per `org-sbe' docstring. I suggest to make an equivalent test
> with, e.g., $"@1$1", where @1$1 refers to a field containing "a\"b\"c"
> or some such.

As I've mentioned, table references are resolved before org-sbe is 
invoked. By the time org-sbe begins executing, all table references will 
have already been replaced by string literals. Using a table reference 
confounds the matter because it is testing more than just org-sbe.

org-sbe's docstring is misleading: $ is not a way to quote just table 
references, but any string literals in general. Had it been otherwise 
(i.e. $-prefixing being part of the table reference syntax), there would 
be no way to pass a string literal (which isn't an interpolated table 
cell value) to the indicated function. The doc string should probably be 
improved in this regard.

Adding a test which uses a table reference instead of a string literal 
won't hurt, but it would be testing several layers at once, and, 
assuming cell value interpolation into emacs lisp table formulas is 
already tested somewhere else, superfluous.

-- 
Best regards,
  Vladimir