How was GCC (GNU Compiler) compiled?

[sha1sum]

Hello guys,

I'm learning the basics of programming and I've got a question. I've read that the program GCC (GNU compiler collection) is written in C++. But what I don't understand is, how did they compile it originally? How can you compile your compiler source code, if you haven't already compiled your compiler?

[Tony Flury]

It was almost certainly compiled/built with an earlier version of the compiler - which in turn was compiled with an earlier compiler .... Eventually you get to the first C++ compiler written in C - go back further, you get the first C compiler (written in Assembler), and even further back you get the first Assembler - written by hand in machine code.

[sha1sum]

I understand. Thanks!

[r-senior]

The process is known as bootstrapping:

https://en.wikipedia.org/wiki/Bootstrapping_(compilers)

[sha1sum]

Thanks for that. Those guys who had to write that first assembler in machine code, must have had a very headache-giving job.

[llanitedave]

There are some crazed geniuses in our past...

[r-senior]

Those guys who had to write that first assembler in machine code, must have had a very headache-giving job.
Actually, writing an assembler in machine code is not a hugely difficult problem because, in its simplest form, an assembler just takes a symbolic representation of machine instructions and translates it into machine code

This is an example of 8080 assembly language from https://en.wikipedia.org/wiki/Intel_8080:


1000 org 1000h ;Origin at 1000h
1000 memcpy public
1000 78 loop mov a,b ;Test BC,
1001 B1 ora c ;If (B or C) = 0,
1002 C8 rz ;Return
1003 1A ldax d ;Load A from (DE)
1004 77 mov m,a ;Store A into (HL)
1005 13 inx d ;Increment DE
1006 23 inx h ;Increment HL
1007 0B dcx b ;Decrement BC
1008 C3 00 10 jmp loop ;Repeat the loop
100B

So "mov a,b" -> 78, "ora c" -> B1 and so on. It's a matter of reading a file, matching simple tokens and writing a file. Obviously non-trivial in machine code, but not mind-bending.

On simple processors, disassemblers are even simpler to write because it's just a matter of looking up numbers in a table and outputting sequences of characters, so 78 -> "mov a,b", B1 -> "ora c" and so on. There's no token matching. Disassemblers probably came before assemblers as a means of visualizing sequences of opcodes.

A much more difficult set of problems comes with a "high-level" programming language: lexical analysis, parsing, code generation and optimization. That was a big leap and spawned tools to help write compilers, like Lex and Yacc. Everything builds on everything else, which is the whole bootstrapping idea.

[ssam]

Thanks for that. Those guys who had to write that first assembler in machine code, must have had a very headache-giving job.

If you are writing a compiler you need to know assembly pretty well anyway.