__FUNCTION__ implemented with mixins and mangles

[Jarrett Billingsley]

2009/6/12 davidl <davidl at nospam.org>:
> 在 Sat, 13 Jun 2009 00:40:09 +0800，Jarrett Billingsley
> <jarrett.billingsley at gmail.com> 写道:
>
>> It's not foolproof, but I found it useful enough; maybe others will too.
>>
>> // Parsing mangles for fun and profit.
>> char[] _getJustName(char[] mangle)
>> {
>>    size_t idx = 1;
>>    size_t start = idx;
>>    size_t len = 0;
>>
>>    while(idx < mangle.length && mangle[idx] >= '0' &&
>>          mangle[idx] <= '9')
>>    {
>>        int size = mangle[idx++] - '0';
>>
>>        while(mangle[idx] >= '0' && mangle[idx] <= '9')
>>            size = (size * 10) + (mangle[idx++] - '0');
>>
>>        start = idx;
>>        len = size;
>>        idx += size;
>>    }
>>
>>    if(start < mangle.length)
>>        return mangle[start .. start + len];
>>    else
>>        return "";
>> }
>>
>> // Eheheh, I has a __FUNCTION__.
>> const char[] FuncNameMix = "static if(!is(typeof(__FUNCTION__)))"
>> "{ struct __FUNCTION {} const char[] __FUNCTION__ ="
>> "_getJustName(__FUNCTION.mangleof); }";
>>
>> To use, just mix into any function where you want to use __FUNCTION__,
>> and it'll be declared as a const char[].
>>
>> void forble()
>> {
>>    mixin(FuncNameMix);
>>    pragma(msg, __FUNCTION__); // shows "forble"
>> }
>>
>> It doesn't seem to cause any noticeable bloat.  The only reference I
>> found to __FUNCTION in an executable compiled with -release was the
>> contents of the FuncNameMix constant itself; I'm sure an "enum string"
>> in D2 wouldn't be emitted like this.
>>
>> It doesn't work for nested functions, but that's just a little more
>> parsing work.  If you want a version that displays the FQN instead of
>> just the function name, I have that too.
>>
>> For those wondering how this works, it's pretty simple: when you
>> declare a type within a function, its mangleof contains the function's
>> name.  All the mixin is doing is declaring a type within the function
>> (struct __FUNCTION), then parsing the owning function's name out of
>> the type's mangleof.
>
> Cool! But basing on mangled name might be broken in the future release of
> DMD.

Only if DMD is generating incorrect mangles now.  Mangles are in the
spec; parsing them is completely valid, cross-platform and
cross-compiler.