Re: [Tinycc-devel] enforced immutability - proposed research project

[Christian Jullien]

Gentlemen,
I'm happy to see you have a lot of ideas. Here is what I propose:
- take time to think about the approach you want to implement this feature, ask 
advices from maintainers if you like.
- move implementation discussion to another list
- THEN propose a research subject with:
  * estimated compilation speed impact (goal/non goal)
  * estimated execution speed impact (goal/non goal)
  * availability/portability on OS/cpu/compiler (IMHO at least mainline 
architectures, I would say Linux/[x64, aarch64, riscv 64], Windows x64 and 
another un*x like *BSD) - Okay, riscv is not yet mainline but hope it will soon.
  * user interface (set using configure option, set using a compiler option, 
new __attribute__ etc...)
  * impact on current source code (mostly implemented in a new file, 
dissimilated in source code with #if/#endif, with macro, ...)
  * list gold ports (i.e. ports tested before any commits) and silver ports 
(i.e. ports that also probably work after commits).
  * nominate a coordinator for this project
- wait approval from maintainers.

Sounds reasonable?

Christian

-----Original Message-----
From: Tinycc-devel [mailto:tinycc-devel-bounces+eligis=orange.fr@nongnu.org] On 
Behalf Of Elijah Stone
Sent: Tuesday, January 19, 2021 02:44
To: tinycc-devel@nongnu.org
Subject: Re: [Tinycc-devel] enforced immutability - proposed research project

On Tue, 19 Jan 2021, Bruno Haible wrote:

> MAP_FIXED is a portability nightmare nowadays, where most OSes implement 
> address space randomization and put the shared libraries at arbitrary 
> and random locations in the address space. You better stay away from it.
>
> The implementation I pointed to therefore doesn't use MAP_FIXED, just 
> plain MAP_SHARED.

The problem is, without a fixed offset between the mutable and immutable 
spaces, certain desirable semantics become harder to implement.  For 
example, we want mutable and immutable references to the same object to 
compare equal.

One potential solution is to rely on overcommit.  Map a giant memory 
region, and then remap the second portion (using MAP_FIXED) to alias the 
first half.  (Using MAP_FIXED to overwrite existing address space mappings 
_is_ a supported usage.)

  -E


_______________________________________________
Tinycc-devel mailing list
Tinycc-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/tinycc-devel