develooper Front page | perl.perl5.porters | Postings from December 2015

Re: Obsolete text in utf8.pm

Thread Previous | Thread Next
From:
John Imrie
Date:
December 24, 2015 18:28
Subject:
Re: Obsolete text in utf8.pm
Message ID:
567C3948.3000603@virginmedia.com
On 23/12/2015 18:54, Karl Williamson wrote:
> On 12/05/2015 05:45 PM, Zefram wrote:
>> Karl Williamson quoted:
>>> One can have UTF-8 in identifier names,
>>
>> Taking "UTF-8" to mean "non-ASCII characters", this is not (and never
>> has been) consistently true.  The de facto identifier syntax varies
>> according to source encoding.  This is one of the remaining instances
>> of the Unicode bug.
>>
>> $ perl -e '$a="\$a\xc0b; 1"; utf8::downgrade($a); eval $a; print $@
>> || "OK\n"'
>> Unrecognized character \xC0; marked by <-- HERE after $a<-- HERE near
>> column 3 at (eval 1) line 1.
>> $ perl -e '$a="\$a\xc0b; 1"; utf8::upgrade($a); eval $a; print $@ ||
>> "OK\n"'
>> OK
>>
>> It is possible to have those characters *if represented in UTF-8 in the
>> source*, but it's not a consistent feature of the language grammar.
>>
>>>                                         but not in package/class or
>>> subroutine names.
>>
>> It was never true that these contexts differed from other kinds of
>> identifier.  From 5.8.0 onwards, up to 5.22, all of these kinds of
>> identifier can have non-ASCII characters if the source is in upgraded
>> form, and cannot if the source is downgraded.
>>
>>>                    While some limited functionality towards this does
>>> exist as of Perl 5.8.0, that is more accidental than designed; use of
>>> UTF-8 for the said purposes is unsupported.
>>
>> Yes, it does look accidental, and we do not in fact support it.
>>
>>> One reason of this unfinishedness is its (currently) inherent
>>> unportability: since both package names and subroutine names may need
>>> to be mapped to file and directory names, the Unicode capability of
>>> the filesystem becomes important-- and there unfortunately aren't
>>> portable answers.
>>
>> This is indeed another issue that has never been addressed and would
>> need to be resolved in order to fully support non-ASCII identifiers.
>> Package names are mapped to pathnames for module loading, and I believe
>> subroutine names for the old system of module splitting.  We essentially
>> need a name mangling layer, to map language-supported identifiers to
>> inoffensive filenames.  (Really, we already need a name mangling layer
>> to cope with case-insensitive filesystems, but in practice we just put
>> up with the resulting misbehaviour.)
>>
>> -zefram
>>
>
> The text in question is referring to under the 'use utf8' pragma.  So
> it really does mean strict UTF-8, and not just non-ASCII.
>
> It appears that UTF-8 is supported in subroutine and package names. 
> How does the following text look as a substitute for what is there now?
>
> BUGS
>
> Some filesystems may not support UTF-8 file names, or they may be
> supported incompatibly with Perl.  Therefore UTF-8 names that are
> visible to the filesystem, such as module names may not work.
>
You have another problem with utf8 in identifiers in that $é is not the
same as $é. Unless something between my email client and your email
client has normalised this message. This could make a difference if you
try to import a sub from a package and the text in the package and
importing script have different ideas of how the identifiers should be
normalised. I would like to propose that identifiers added into the
stash by Perl its self are done so in Normalization Form C this does not
stop a programmer from directly manipulating the stash with such as
${"é::é"} but in this case they are supposed to know what they are doing.

Thoughts please.

Thread Previous | Thread Next


nntp.perl.org: Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at ask@perl.org | Group listing | About