#279 limit shimmer of singleton lists

[Don Porter]

If a Tcl_Obj is a single value, we lose it's intrep
when converting to the "list" objType.

We can avoid loss of intrep by just making the
object into the first element of the objPtr
array of the List struct, so long as we're careful
about string reps that would be broken into
multiple elements.

Here's a patch that almost works. It makes
test append-4.8 fail. Comments?

[Don Porter]

[Kevin B KENNY]

Logged In: YES
user_id=99768
Originator: NO

Is this still an issue after so long? (I know that you have been in the code a good bit more recently than I have.)

Don's patch no longer applies.

Avoiding shimmer of singletons sounds like a good idea, if we can make it go.

[miguel sofer]

Logged In: YES
user_id=148712
Originator: NO

Attached a current patch with three shortcuts:
- it avoids most work for empty strings
- it avoids computing the string rep for numbers
- it avoids losing the internal rep for singletons that do not have a string rep

I am a bit mystified as to why the qualification about not having a string rep is necessary; without it, failures at: append-4.8, appendComp-4.8, opt-[3.1/3.2/5.1], safe-7.2

File Added: listDiff

[miguel sofer]

[miguel sofer]

Logged In: YES
user_id=148712
Originator: NO

New patch after first comments
File Added: listDiff

[miguel sofer]

[miguel sofer]

Logged In: YES
user_id=148712
Originator: NO

Mystery solved: a string that contains backslashes or braces needs full parsing. Updated patch, committed.
File Added: listDiff2

[Jeffrey Hobbs]

Logged In: YES
user_id=72656
Originator: NO

This patch causes Tk's border-1.1, bitmap-1.1 and cursor-1.1 to fail.

[miguel sofer]

Logged In: YES
user_id=148712
Originator: NO

* generic/tclListObj.c: reverting [Patch 738900] (committed on
2007-04-20). Causes some Tk test breakage of unknown importance,
but the impact of the patch itself is likely to be so small that
it does not warrant investigation at this time.