Unexpected issue with std.format

[Saurabh Das]

On Saturday, 1 February 2020 at 15:16:41 UTC, Steven 
Schveighoffer wrote:
> On 2/1/20 8:39 AM, Saurabh Das wrote:
>> I faced this issue while working with custom formatting for a 
>> struct. I have reduced the error down to this test program:
>> 
>> import std.format, std.stdio, std.array;
>> 
>> struct Test1
>> {
>>      void toString(W, C)(ref W w, scope const ref FormatSpec!C 
>> fmt)
>>      {
>>          pragma(msg, "Test1 function compiled with W=" ~ 
>> W.stringof);
>>          // formatValue(w, this, fmt);
>>      }
>> }
>> 
>> struct Test2
>> {
>>      void toString(W, C)(ref W w, scope const ref FormatSpec!C 
>> fmt)
>>      {
>>          pragma(msg, "Test2 function compiled with W=" ~ 
>> W.stringof);
>>          formatValue(w, this, fmt);
>>      }
>> }
>> 
>> void main()
>> {
>>      Test1 t1;
>>      Test2 t2;
>> 
>>      Appender!string writer;
>>      auto ff = singleSpec("%s");
>> 
>>      formatValue(writer, t1, ff);
>>      formatValue(writer, t2, ff);
>> }
>> 
>> When compiled, the output is:
>> 
>> Test1 function compiled with W=S
>> Test1 function compiled with W=Appender!string
>> Test2 function compiled with W=S
>> 
>> 1. Why was Test2 never compiled with W=Appender!string?
>> 2. What is "S"?
>> 
>> Essentially, my custom struct was not being formatted using 
>> the toString method that I had written. Reducing the issue, it 
>> seems like a call to formatValue with the same type caused the 
>> issue. If someone can explain what I am doing wrong here, it 
>> would really help a lot.
>> 
>> Thanks,
>> Saurabh
>> 
>
> Something very weird is happening.
>
> I switched to fullyQualifiedName!W, and I get *no output*.
>
> The "S" comes from hasToString 
> here:https://github.com/dlang/phobos/blob/9fe5cd354f0166b11d32a5c1214932757d8e7eba/std/format.d#L3876-L3899
>
> I tried copying the implementation to a local file, and as 
> expected, I get customPutWriterFormatSpec for both types.
>
> But it's only calling one of them in the implementation.
>
> I think the only true way to diagnose this is to instrument 
> std.format and see what it's doing with more pragma(msg) calls. 
> Don't have the time right now.
>
> -Steve

Thanks for the lead. To exemplify Steve's observation:

import std.format, std.stdio, std.array, std.range;

struct Test3
{
     void toString(W, C)(ref W w,  scope const ref FormatSpec!C 
fmt)
     {
         import std.traits;
         pragma(msg, "A: Test2 function compiled with W=" ~ 
fullyQualifiedName!W.stringof);
         pragma(msg, "B: Test2 function compiled with W=" ~ 
W.stringof);
     }
}

void main()
{
     Test3 t3;

     Appender!string writer;
     auto ff = singleSpec("%s");

     formatValue(writer, t3, ff);
}

Gives an output during compilation:
B: Test2 function compiled with W=S


Saurabh