name mangling: bring documentation closer to actual implementation

[rainers]

This documents what I found in the current implementation. Not everything seems desirable, though...

[MartinNowak]

LGTM

[MartinNowak]

Let's move the less common anonymous symbol case to the bottom.

[rainers]

Amended.

BTW: I've only seen this with overload sets as template arguments, but it looks like a bug to me: the overloaded symbol has these anonymous parents, but something like the location of the lookup should be encoded instead.

[wilzbach]

@rainers: how about linking to the bison spec file that you curated? I imagine that this could help other people as well (at least those reading the spec about name mangling)

[rainers]

@wilzbach A bison grammar matching the documentation is now here: https://gist.githubusercontent.com/rainers/6cdf73b48837defb9f88/raw/0028d4aed82b0abf2a5a32322398ebe4595903bb/dwebsite3.bison

Not sure where to place and link to it. I think it would have to be part of the dlang.org repository to have a chance of staying in sync.

BTW: It shows 2 ambiguities in the grammar:

• 'M' can be scope both on the function itself or the first function argument.
• Y can be the end of an function argument list or the beginning of the type of an Objective-C function argument.

[wilzbach]

Not sure where to place and link to it. I think it would have to be part of the dlang.org repository to have a chance of staying in sync.

How about creating a new data or bison folder?
What is your opinion @CyberShadow?

[CyberShadow]

IMO? Put it in the same directory as the name mangling spec. Also weird, can't comment via the web interface: https://dump.thecybershadow.net/3b56e04862e0c61b7e3135ac95d3b129/13%3A14%3A06-upload.png

…

On 2016-12-19 08:30, Sebastian Wilzbach wrote: > Not sure where to place and link to it. I think it would have to be part of the dlang.org repository to have a chance of staying in sync. How about creating a new `data` or `bison` folder? What is your opinion @CyberShadow?

-- Best regards, Vladimir

[ibuclaw]

Actually, this should be QualifiedName / MangledName.

Example: https://github.com/dlang/phobos/blob/31e1b462d2311f9d8cecb33172b28eab587b255d/std/parallelism.d#L43-L77

Instantiates this symbol:

std.parallelism.Task!(run, real delegate(MapResult!(getTerm, Result), ulong, ulong) @system, MapResult!(getTerm, Result), ulong, ulong).Task.impl

Mangled as:

_D3std11parallelism301__T4TaskS213std11parallelism3runTDFS3std9algorithm9iteration91__T9MapResultS31_D8parallel4mainFZ7getTermMFiZeTS3std5range13__T4iotaTiTiZ4iotaFiiZ6ResultZ9MapResultmmZeTS3std9algorithm9iteration91__T9MapResultS31_D8parallel4mainFZ7getTermMFiZeTS3std5range13__T4iotaTiTiZ4iotaFiiZ6ResultZ9MapResultTmTmZ4Task4implFPvZv

Right at the beginning of the first template you have:

__T4TaskS213std11parallelism3run
           ^

Also, this cunningly forgets Bugzilla 3043 where symbol arguments are prefixed with their length.

The correct grammar should be:

         $(GNAME TemplateArgX):
            $(B T) $(GLINK Type)
            $(B V) $(GLINK Type) $(GLINK Value)
            $(B S) $(GLINK Number) $(GLINK QualifiedName)
            $(B S) $(GLINK Number) $(GLINK MangledName)

[rainers]

Yeah, noticed this too last night, see https://issues.dlang.org/show_bug.cgi?id=14591#c7