[phobos] Deprecation of std.regexp

[Dmitry Olshansky]

On 06.06.2011 3:20, Jonathan M Davis wrote:
> On 2011-06-05 09:38, Dmitry Olshansky wrote:
>> On 05.06.2011 12:13, Jonathan M Davis wrote:
>>> On 2011-06-05 00:56, Brad Roberts wrote:
>>>> On 6/5/2011 12:52 AM, Jonathan M Davis wrote:
>>>>> It looks like std.regexp was marked in its documentation as deprecated
>>>>> in 2.053. It was not actually deprecated (the deprecation modifier is
>>>>> missing), but if it has indeed been deprecated (and std.regex has been
>>>>> around for a while, so it's not exactly a surprise), then it needs to
>>>>> actually be deprecated, and the rest of Phobos needs to be using
>>>>> std.regex instead. I can easily deprecate the function (after all, all
>>>>> you have to do is add deprecated: near the top of the file), but since
>>>>> I've never used either of the regex modules, I'm ill-suited to convert
>>>>> the rest of Phobos to use std.regex. So, I'd appreciate it if someone
>>>>> who is actually familiar with the two modules would convert the rest
>>>>> of Phobos to use std.regex sometime prior to the next release, and
>>>>> then we can actually deprecate std.regexp. Worse comes to worst, I can
>>>>> take a crack at it, but it would be much faster if someone who's
>>>>> familiar with the regex modules did it.
>>>>>
>>>>> Now, we may actually want to change std.regexp to "scheduled for
>>>>> deprecation" rather than deprecated simply because there are other
>>>>> functions in Phobos which take stuff from std.regexp but not std.regex
>>>>> (such as the version of std.file.listdir which takes a RegExp) have not
>>>>> yet been scheduled for deprecation, let alone deprecated, but
>>>>> regardless, we need to change the rest of Phobos to use std.regex or we
>>>>> won't be able to actually deprecate and remove std.regexp (well, I
>>>>> suppose that we _could_ deprecate it as-is, but it seems to me to be a
>>>>> bad idea to deprecate something when Phobos is still using it in its
>>>>> API elsewhere).
>>>>>
>>>>> - Jonathan M Davis
>>>> It was marked deprecated once (check out the blame history of the file),
>>>> but that was reverted since commonly used modules still use it.  The
>>>> effect would be that essentially every app would need to be compiled
>>>> with -d, which is hardly useful.  Until phobos doesn't use it, no real
>>>> deprecation can actually occur.
>>> That's essentially my point. Someone needs to convert the rest of Phobos
>>> so that it uses std.regex instead of std.regexp before we can properly
>>> deprecate std.regexp (and given all of the stuff in Phobos which still
>>> uses std.regexp, we should probably make it so that std.regexp is
>>> _scheduled_ for deprecation rather than deprecated, but the changes
>>> still need to be made). Simply marking std.regexp as deprecated in its
>>> documentation isn't enough. I'm raising the issue so that someone who is
>>> actually familiar with std.regexp and std.regex can do the conversion.
>> Looks like I'm the one best suited for this, since I have extensive
>> experience with both modules and their internals.
>> In fact, I'm doing GSOC project aimed at enhancing current std.regex,
>> basically replacing implementation and extending interface quite a bit.
>> Still changes going to be backwards compatible, so it's about time to
>> switch all phobos to std.regex.
>> I'll make a pull request soon.
> Thanks! Remember that where possible, functions in Phobos' public API which
> would be affected by the change should have the old version kept temporarily
> as "scheduled for deprecation" in order to give people some time to convert
> over rather than immediately breaking their code. And if the old function is a
> template, then a pragma warning about the the change should be put inside it.
> You can look at what std.file is currently doing with functions scheduled for
> deprecation for examples. Don't worry too much about the exact messages. I can
> go and adjust them later. Just mark them as scheduled for deprecation.
>
> The current plan is that anything which is deprecated should be scheduled for
> deprecation for about 6 months, deprecated for 6 months, and then removed. But
> std.regex has been around long enough that I suspect that we can accelerate
> that process a bit in this case.
>
> In any case, thank you for taking the time to convert over Phobos to
> std.regex. Then we can finally move towards removing std.regexp.

It turns out far easier then expected but at the same time far nastier. 
There is *only one* function in phobos that use std.regexp it's listDir 
in std.file. Now I just provide another overload for it and boom  - 
forward reference bug:

std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(326): Error: template std.conv.text(T...) cannot deduce 
template function from argument types !()(string,uint,string)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(327): Error: template std.conv.text(T...) cannot deduce 
template function from argument types !()(string,uint)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(328): Error: template std.conv.text(T...) cannot deduce 
template function from argument types 
!()(string,_error_,string,_error_,string)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(331): Error: template std.conv.text(T...) cannot deduce 
template function from argument types 
!()(string,_error_,string,const(immutable(char)[]),string)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(326): Error: template std.conv.text(T...) cannot deduce 
template function from argument types !()(string,uint,string)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(327): Error: template std.conv.text(T...) cannot deduce 
template function from argument types !()(string,uint)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(328): Error: template std.conv.text(T...) cannot deduce 
template function from argument types 
!()(string,_error_,string,_error_,string)
std\conv.d(3762): Error: template std.conv.text(T...) forward reference 
to template text(T...)
std\typecons.d(331): Error: template std.conv.text(T...) cannot deduce 
template function from argument types 
!()(string,_error_,string,const(immutable(char)[]),string)
std\typecons.d(364): Error: cannot evaluate injectNamedFields() at 
compile time
std\typecons.d(364): Error: argument to mixin must be a string, not 
(injectNamedFields())
std\regex.d(185): Error: template instance 
std.typecons.Tuple!(uint,"startIdx",uint,"endIdx") error instantiating
std\file.d(3421):        instantiated from here: Regex!(char)


It works outside of std.file, so I'm not sure what to try next.
To reproduce just add std.regex to imports in std.file. And add this as 
another overload for listDir.
/++ Ditto +/

string[] listDir(C)(in C[] pathname, Regex!char r, bool followSymLinks = 
true)
{
     auto result = appender!(string[])();

     bool callback(DirEntry* de)
     {
         if(followSymLinks ? de.isDir : isDir(de.linkAttributes))
         {
             _listDir(de.name, &callback);
         }
         else if(true)//(!match(de.name,r).empty)
         {
             result.put(de.name);
         }

         return true; // continue
     }

     _listDir(pathname, &callback);

     return result.data;
}

> - Jonathan M Davis
> _______________________________________________
> phobos mailing list
> phobos at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/phobos


-- 
Dmitry Olshansky