如何结合不同的编程语言

Question

I'm not asking about WHEN to link different programming langauges.

This is quite a general question but I'm personally working on Linux.

What I want to understand is the process by which different programming languages can be combined, I found a good article on combining C/C++/Fortran: http://www-h.eng.cam.ac.uk/help/tpl/languages/mixinglanguages.html .

From what I understand most compilers perform two stages:

1. Translating the language files into object files which contain machine code but still contain some symbols (possibly function names?)

2. Linking the object files together, only at this stage the Linker checks that the functions in the object files are callable.

I think that the problem with combining different languages is name mangling which means that the names of the functions are changed when they are turned into object code.

The questions are:

1. Can't you somehow discover the mangled function names beforehand and than specify them explicitly in the programming language or better off, isn't there a software that already does that?

2. I don't understand completely how dynamic libraries are linked but can different languages interact by the same method that programs interact with dynamic libraries?

ps The main intent is to call functions written in another language.

Answer 1

The issue with linking different object files together generally comes down to subroutine calling conventions. Basically, when you make a call to a routine located in another object file, your compiler will have to know what that other object file will name its routine internally, how to pass all its parameters, and what (if any) setup and cleanup code the routine will require. All this stuff is generally grouped together under the heading of calling convention s.

Each compiler has its own calling conventions it likes to use for subroutines. Note I said "compiler", not language. The C calling convention in Linux is different than the C calling convention on Windows.

So when you mix languages, you need some way to tell the compiler for either the calling or the called subroutine to use the other language's calling convention. C's convention is a popular one to use as sort of a "lingua franca", as just about every platform has a C compiler. However some platforms (eg: Windows) have multiple popular calling conventions.

So now we ask the question you asked in the comments:

Is there a common way to "tell the compiler to use the other language's calling convention"?

And the answer is, "No, not really". Some languages do have defined ways of using specific other language's calling conventions. For example, C++ allows you to to put extern "C" on declarations to tell the compiler that the declaration(s) in question use the C calling convention. Ada accomplishes the same thing with pragma Convention (X,...) , where X is the convention name. C , Fortran , and Cobol are defined by the language, but anything else supported (eg: Windows' Stdcall ) is implementation defined.

However, if you have a pair of languages whose compiler writers never thought of each other, then you have no choice but to tell both to use some third convention that they both know about (usually C's). For example, to get standard C++ and Ada to interoperate, you'd have the server code export its routines using the C convention, and tell the client code that the routines it is calling are using the C convention.

Answer 2

Different languages can definitely use the same libraries. On the old Windows Visual Basic it was quite common to dynamically load Windows API functions, for instance.

All you need for inter-language linking is an agreement on the function's calling conventions, along with knowledge of the function names. The former has to be done by looking up the documentation; the latter has to be looked up in the compiler that created the objects or libraries. For example, gcc will compile C without mangling names, so you can refer directly to the function names as they are in your C source, while g++ will compile C++ code with mangled names and you're best off exposing C functions via extern "C" declarations.

Basically, as long as your objects or libraries expose only the C ABI, there should be widespread support for binding to other languages. It's a lot more difficult if you want to use a native C++ library, for instance, since in that case your foreign languages have to implement the correct C++ ABI. It's similar for exporting code from, say, Fortran, but I believe that one can be made to just uses the C ABI.

Answer 3

The "standard" is to use non-mangled names when combining programs from different languages. Name mangling can be turned off for specific symbols in C++ by declaring them with extern "C" . C does not mangle names.

Answer 4

All library executables contain some type of interface. If they did not, no software would be able to work with them. It is more likely internal methods get changed to be more efficient. In addition, many languages allow you to turn off "mangling" at the compiler level.

Linking, as a simple explanation (I will probably get dinked for this?), is packaging into a single file. The classes retain the same interface as non-linked libraries, at least from an external programming standpoint.