Project Logs

Collapse

• anti-break do()

  Yann Guidon / YGDES • 04/12/2025 at 00:27 • 0 comments

  I added an inner loop inside the DS_MARK-DS_RELEASE macro :
  

  // The inseparable pair :
  #define DS_MARK                      \
  do {                                 \
    datastack_mark_type mark_variable; \
    DO_MARK(mark_variable);            \
    do {
  
  // double-loop prevents break from skipping the release
  
  #define DS_RELEASE                            \
    } while(0);                                 \
    scan_destructors(mark_variable.destructor); \
    DO_RELEASE(mark_variable);                  \
  } while(0);

  It only occurred to me that a break between the two macros would skip the release... now it's fixed.
  

• Use-after-free

  Yann Guidon / YGDES • 03/05/2025 at 00:32 • 0 comments

  I just dropped stack_alloc.tgz
  

  This addresses a big concern, the hard way : you can clear a pointer after the corresponding block has been freed by RELEASE. So if you use this pointer after, you SEGFAULT.

  Of course, this pointer should not have been copied elsewhere because there is no reference count but at least you have an "early safety blanket" in case of bug : fail early and find out why before it's too intricate.

  From example_clearptr.c:

    // How it should ideally be done:
    DS_MARK
      puts("1st version: should work smoothly.");
      char *ch1=new(32);
      memcpy(ch1, alphabet, 27);
      puts(ch1);
    DS_RELEASE
    // ch1 is out of declaration scope here and
    // can't be used after DS_RELEASE freed it.
  
    // Less ideal version where the pointer gets
    // NULLed by the destructor to prevent use-after-free
    char *ch2;
    DS_MARK
      puts("2nd version: must SEGFAULT:");
      new_clear(32, ch2);
      memcpy(ch2, alphabet+3, 27-3);
      // puts must work here
    DS_RELEASE
    // puts shouldn't work here because the pointer has been freed.
    puts(ch2);
  

  It's quite brutal but it works without being overkill !
  

• First archive

  Yann Guidon / YGDES • 03/03/2025 at 17:40 • 0 comments

  The core code uses about 330 lines (.c+.h) and can now be found, without the aStrA part, there :
  

  data_stack.tgz
  

  It supports three options:

  • DATA_STACK_ADJUSTABLE : allows the top object to be grown or shrunk after new().
  • DATA_STACK_DESTRUCTOR : an "object" can be associated with a function pointer, called in reverse order "like a local atexit()"
  • DATA_STACK_MAXALLOC : collect maximum stack depth/usage.

  Two example programs are provided.
  

• First code

  Yann Guidon / YGDES • 12/28/2024 at 08:56 • 0 comments

  So I was working on the #aStrA : Aligned Strings format with attributesand the principles have resurfaced.

  I also write an article on the subject, the current dev snapshot is at AS20241228.tgz.

  It's still a dumb mark-new-release allocator but it is multithread-compatible, you can locate the buffer anywhere you like in data memory.
  

• First article

  Yann Guidon / YGDES • 05/08/2024 at 00:52 • 0 comments

  A first article is now available in the French new stands in May-June 2024:

  

  It's only the first part of a trilogy that explains many of the practical shortcomings of the C-style ABI that makes our lives so miserable. Beware of the C hacks...
  

View all 5 project logs