bug#37820: 25.2; xref using etags incorrect case fold during completion

[Radey Shouman <shouman@comcast.net>]

Dmitry Gutov writes:
 > Hi Radey,
 > 
 > sorry for the late reply.

Not at all.

 > On 25.10.2019 16:59, Radey Shouman wrote:
 > > The only variable I customize is tags-case-fold-search, I leave
 > > completion-ignore-case with the default value.  I heartily regret
 > > even mentioning the latter variable, it is a red herring.
 > 
 > Don't regret it because it's entirely pertinent.
 > 
 > So, do we really have meaningful scenarios where the user will want 
 > tags-case-fold-search set to t while completion-ignore-case *should* 
 > remain nil?

No.  I believe that completion-ignore-case should always be bound to
the value of tags-case-fold-search when doing a tags completion.  If
this is not so then completion is not consistent with the tags search,
which is the bug in question.

 > I suppose we could add a new method to xref backends just for that, but 
 > we shouldn't do that without understanding the use cases. Like, why/how 
 > other backends would choose to return different values.
 > 
 > > There are at least two reasons to change tags-case-fold-search:
 > > 
 > > 1) Some programming languages are not case sensitive in normal use,
 > > eg Common Lisp, Fortran, Cobol, R[45]RS Scheme ...  For these languages
 > > case-insensitive search should be the default.
 > 
 > How does using one or the other variable help with that distinction?

From what I could see searching the lisp tree, tags-case-fold-search
is intended for user customization of tags search (and partly carried
over to xref).  completion-ignore-case is used for dynamic control of
completion, and is not a sensible variable for user customization.

 > When someone uses tags, they often use them with a multitude of 
 > languages. Whether you employ tags-case-fold-search, or an xref-specific 
 > variable Eli suggested to add, that would affect all languages anyway, 
 > right?

Yes.  It is possible to mix languages even in a single tags table.
Personal preference also plays a role.

 > > 2) For many code bases the rules of CamelCasing are so vague, or their
 > > observance so lax, that it's hard to predict what the capitalization
 > > of an identifier will be.  It's useful in these cases to use
 > > completion that is not case sensitive.
 > 
 > We can make completion-ignore-case a safe local variable, so that you 
 > can set it in dir-locals depending on the code base.

I'm not sure that's useful, because the default value of
completion-ignore-case is hardly ever used.  For example, even
if ignoring case for tags searches, one would probably not want to
ignore case when completing input for describe-key.

On the other hand, it might well be useful to make
tags-case-fold-search or any xref replacement a safe local variable.

BTW, how might one find the list of safe local variables?

--