R devel - May 2020 - Minor Infelicity in Printing of Objects Nested in Lists

Currently S3 objects nested in generic vectors cause the tag buffer to be
reset.? This feels sub-optimal for those objects that don't have a print
method:
> list(a=list(b='hello'))
$a
$a$b???????????????? ### <<<< notice "$a$b"
[1] "hello"

> list(a=structure(list(b='hello'), class='world'))
$a
$b?????????????????? ### <<<< notice "$b", not
"$a$b"
[1] "hello"

attr(,"class")
[1] "world"

This happens because the default print method resets the tag buffer anytime
it's called[1], whether by a custom print method, or by internal C code as
part of the recursion into objects in `printValueRec`[2].

One possible way to "fix" this is to make it the responsibility of
`printValueRec` to reset the tag buffer on exit from the top level, but that
would mean always having an active context to catch errors instead of just for
win32 as is the case now.? Additionally, print method authors may themselves
intentionally rely on `print.default` for some parts their output, in which case
it may be desirable to reset the tag buffer.

Obviously not a big deal.? In my use case it can make it difficult to identify
what part of a complex object caused a difference in a diff with a few
neighboring context lines.? Just mentioning it in case there is interest in
doing something about it.? If so I'll be happy to create a bugzilla ticket
and assist with patches/testing.

Best,

B.

[1]: https://github.com/wch/r-source/blob/tags/R-4-0-0/src/main/print.c#L309
[2]: https://github.com/wch/r-source/blob/tags/R-4-0-0/src/main/print.c#L570

Hello,

The main reason for resetting the tagbuf in `print.default()` and
other entry points to the print routine is that it is currently not
reset on exit. Creating a context to reset it on exit to its last
value might work. This should be done in the entry points rather than
in print-value-rec though, since callers of the latter might write to
the tagbuf.

Another solution to this problem is proposed with the first patch in
https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17610. Maybe I
should extract this patch into its own entry so the discussion can be
separate from the other patches.

The idea is to pass the tag buffer as argument to the `print()`
generic so it can be passed again to `print.default()`. The added
benefit is that print methods can then pass their own tagbuf, which
should be useful for printing special list-like objects.

The downside is that it is technically a breaking change for methods
that pass their dots to other functions (there's a couple of
occurrences of this in the base packages). However I think such usage
is not recommended for print methods because `print()` is a recursive
generic that dispatches on heterogeneous objects. For example the user
might pass `digits = 2` to print a list of objects. The print methods
should make sure they ignore any unknown arguments.

Best,
Lionel

On 5/10/20, brodie gaslam via R-devel <r-devel at r-project.org>
wrote:
> Currently S3 objects nested in generic vectors cause the tag buffer to be
> reset.? This feels sub-optimal for those objects that don't have a
print
> method:
>
>> list(a=list(b='hello'))
> $a
> $a$b???????????????? ### <<<< notice "$a$b"
> [1] "hello"
>
>
>> list(a=structure(list(b='hello'), class='world'))
> $a
> $b?????????????????? ### <<<< notice "$b", not
"$a$b"
> [1] "hello"
>
> attr(,"class")
> [1] "world"
>
> This happens because the default print method resets the tag buffer anytime
> it's called[1], whether by a custom print method, or by internal C code
as
> part of the recursion into objects in `printValueRec`[2].
>
> One possible way to "fix" this is to make it the responsibility
of
> `printValueRec` to reset the tag buffer on exit from the top level, but
that
> would mean always having an active context to catch errors instead of just
> for win32 as is the case now.? Additionally, print method authors may
> themselves intentionally rely on `print.default` for some parts their
> output, in which case it may be desirable to reset the tag buffer.
>
> Obviously not a big deal.? In my use case it can make it difficult to
> identify what part of a complex object caused a difference in a diff with a
> few neighboring context lines.? Just mentioning it in case there is
interest
> in doing something about it.? If so I'll be happy to create a bugzilla
> ticket and assist with patches/testing.
>
> Best,
>
> B.
>
> [1]:
> https://github.com/wch/r-source/blob/tags/R-4-0-0/src/main/print.c#L309
> [2]:
> https://github.com/wch/r-source/blob/tags/R-4-0-0/src/main/print.c#L570
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

> On Sunday, May 10, 2020, 12:24:17 PM EDT, Lionel Henry <lionel at
rstudio.com> wrote:
>
> The main reason for resetting the tagbuf in `print.default()` and
> other entry points to the print routine is that it is currently not
> reset on exit. Creating a context to reset it on exit to its last
> value might work. This should be done in the entry points rather than
> in print-value-rec though, since callers of the latter might write to
> the tagbuf.
>
> Another solution to this problem is proposed with the first patch in
> https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17610. Maybe I
> should extract this patch into its own entry so the discussion can be
> separate from the other patches.
Ah, looks like the exact same issue, at least the first patch.? I'm not
 wed to any particular solution, but I am happy to help test/work on this
if there is interest in addressing this.

Best,

B.
>On 5/10/20, brodie gaslam via R-devel <r-devel at r-project.org>
wrote:
>> Currently S3 objects nested in generic vectors cause the tag buffer to
be
>> reset.? This feels sub-optimal for those objects that don't have a
print
>> method:
>>
>>> list(a=list(b='hello'))
>> $a
>> $a$b???????????????? ### <<<< notice "$a$b"
>> [1] "hello"
>>
>>
>>> list(a=structure(list(b='hello'), class='world'))
>> $a
>> $b?????????????????? ### <<<< notice "$b", not
"$a$b"
>> [1] "hello"
>>
>> attr(,"class")
>> [1] "world"