diff mbox series

t-reftable-basics: allow for `malloc` to be `#define`d

Message ID pull.1848.git.1736352005578.gitgitgadget@gmail.com (mailing list archive)
State Accepted
Commit d02c37c3e6baf1515e7d1372afa5941b9518ca5b
Headers show
Series t-reftable-basics: allow for `malloc` to be `#define`d | expand

Commit Message

Johannes Schindelin Jan. 8, 2025, 4 p.m. UTC
From: Johannes Schindelin <johannes.schindelin@gmx.de>

As indicated by the `#undef malloc` line in `reftable/basics.h`, it is
quite common to use allocators other than the default one by defining
`malloc` constants and friends.

This pattern is used e.g. in Git for Windows, which uses the powerful
and performant `mimalloc` allocator.

Furthermore, in `reftable/basics.c` this `#undef malloc` is
_specifically_ disabled by virtue of defining the
`REFTABLE_ALLOW_BANNED_ALLOCATORS` constant before including
`reftable/basic.h`, to ensure that such a custom allocator is also used
in the reftable code.

However, in 8db127d43f5b (reftable: avoid leaks on realloc error,
2024-12-28) and in 2cca185e8517 (reftable: fix allocation count on
realloc error, 2024-12-28), `reftable_set_alloc()` function calls were
introduced that pass `malloc`, `realloc` and `free` function pointers as
parameters _after_ `reftable/basics.h` ensured that they were no longer
`#define`d. This would override the custom allocator and re-set it to
the default allocator provided by, say, libc or MSVCRT.

This causes problems because those calls happen after the initial
allocator has already been used to initialize an array, which is
subsequently resized using the overridden default `realloc()` allocator.

You cannot mix and match allocators like that, which leads to a
`STATUS_HEAP_CORRUPTION` (C0000374) on Windows, and when running this
unit test through shell and/or `prove` (which only support 7-bit status
codes), it surfaces as exit code 127.

It is actually unnecessary to use those function pointers to
`malloc`/`realloc`/`free`, though: The `reftable` code goes out of its
way to fall back to the initial allocator when passing `NULL` parameters
instead. So let's do that instead of causing heap corruptions.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
---
    t-reftable-basics: allow for malloc to be #defined
    
    This is a fix for one of the many issues that force me to delay Git for
    Windows v2.48.0-rc2 until I can increase my confidence via thorough
    testing.
    
    The patch is based on rs/reftable-realloc-errors. Sadly, the patch fails
    the PR build
    [https://github.com/gitgitgadget/git/actions/runs/12672507500/job/35316720255],
    but then the base branch fails in the same way
    [https://github.com/gitgitgadget/git/actions/runs/12533205564/job/34952668803].

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1848%2Fdscho%2Freftable-tests-should-allow-malloc-to-be-defined-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1848/dscho/reftable-tests-should-allow-malloc-to-be-defined-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1848

 t/unit-tests/t-reftable-basics.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)


base-commit: 1e781209284eb5952e153339f45bf0c1555e78bb

Comments

René Scharfe Jan. 8, 2025, 4:36 p.m. UTC | #1
Am 08.01.25 um 17:00 schrieb Johannes Schindelin via GitGitGadget:
> From: Johannes Schindelin <johannes.schindelin@gmx.de>
>
> As indicated by the `#undef malloc` line in `reftable/basics.h`, it is
> quite common to use allocators other than the default one by defining
> `malloc` constants and friends.
>
> This pattern is used e.g. in Git for Windows, which uses the powerful
> and performant `mimalloc` allocator.
>
> Furthermore, in `reftable/basics.c` this `#undef malloc` is
> _specifically_ disabled by virtue of defining the
> `REFTABLE_ALLOW_BANNED_ALLOCATORS` constant before including
> `reftable/basic.h`, to ensure that such a custom allocator is also used
> in the reftable code.
>
> However, in 8db127d43f5b (reftable: avoid leaks on realloc error,
> 2024-12-28) and in 2cca185e8517 (reftable: fix allocation count on
> realloc error, 2024-12-28), `reftable_set_alloc()` function calls were
> introduced that pass `malloc`, `realloc` and `free` function pointers as
> parameters _after_ `reftable/basics.h` ensured that they were no longer
> `#define`d. This would override the custom allocator and re-set it to
> the default allocator provided by, say, libc or MSVCRT.
>
> This causes problems because those calls happen after the initial
> allocator has already been used to initialize an array, which is
> subsequently resized using the overridden default `realloc()` allocator.
>
> You cannot mix and match allocators like that, which leads to a
> `STATUS_HEAP_CORRUPTION` (C0000374) on Windows, and when running this
> unit test through shell and/or `prove` (which only support 7-bit status
> codes), it surfaces as exit code 127.
>
> It is actually unnecessary to use those function pointers to
> `malloc`/`realloc`/`free`, though: The `reftable` code goes out of its
> way to fall back to the initial allocator when passing `NULL` parameters
> instead. So let's do that instead of causing heap corruptions.

Ugh.  That makes a lot of sense.  Sorry for the trouble! :-/

>
> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> ---
>     t-reftable-basics: allow for malloc to be #defined
>
>     This is a fix for one of the many issues that force me to delay Git for
>     Windows v2.48.0-rc2 until I can increase my confidence via thorough
>     testing.
>
>     The patch is based on rs/reftable-realloc-errors. Sadly, the patch fails
>     the PR build
>     [https://github.com/gitgitgadget/git/actions/runs/12672507500/job/35316720255],
>     but then the base branch fails in the same way
>     [https://github.com/gitgitgadget/git/actions/runs/12533205564/job/34952668803].
>
> Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1848%2Fdscho%2Freftable-tests-should-allow-malloc-to-be-defined-v1
> Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1848/dscho/reftable-tests-should-allow-malloc-to-be-defined-v1
> Pull-Request: https://github.com/gitgitgadget/git/pull/1848
>
>  t/unit-tests/t-reftable-basics.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/t/unit-tests/t-reftable-basics.c b/t/unit-tests/t-reftable-basics.c
> index 990dc1a2445..1d640b280f9 100644
> --- a/t/unit-tests/t-reftable-basics.c
> +++ b/t/unit-tests/t-reftable-basics.c
> @@ -157,13 +157,13 @@ int cmd_main(int argc UNUSED, const char *argv[] UNUSED)
>
>  		old_alloc = alloc;
>  		old_arr = arr;
> -		reftable_set_alloc(malloc, realloc_stub, free);
> +		reftable_set_alloc(NULL, realloc_stub, NULL);
>  		check(REFTABLE_ALLOC_GROW(arr, old_alloc + 1, alloc));
>  		check(arr == old_arr);
>  		check_uint(alloc, ==, old_alloc);
>
>  		old_alloc = alloc;
> -		reftable_set_alloc(malloc, realloc, free);
> +		reftable_set_alloc(NULL, NULL, NULL);
>  		check(!REFTABLE_ALLOC_GROW(arr, old_alloc + 1, alloc));
>  		check(arr != NULL);
>  		check_uint(alloc, >, old_alloc);
> @@ -188,11 +188,11 @@ int cmd_main(int argc UNUSED, const char *argv[] UNUSED)
>  		arr[alloc - 1] = 42;
>
>  		old_alloc = alloc;
> -		reftable_set_alloc(malloc, realloc_stub, free);
> +		reftable_set_alloc(NULL, realloc_stub, NULL);
>  		REFTABLE_ALLOC_GROW_OR_NULL(arr, old_alloc + 1, alloc);
>  		check(arr == NULL);
>  		check_uint(alloc, ==, 0);
> -		reftable_set_alloc(malloc, realloc, free);
> +		reftable_set_alloc(NULL, NULL, NULL);

Using NULL also captures the intent to set the default allocator better.

>
>  		reftable_free(arr);
>  	}
>
> base-commit: 1e781209284eb5952e153339f45bf0c1555e78bb
Junio C Hamano Jan. 8, 2025, 5:14 p.m. UTC | #2
"Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
writes:

> -		reftable_set_alloc(malloc, realloc_stub, free);
> +		reftable_set_alloc(NULL, realloc_stub, NULL);

Nice.  By setting it to NULL, we force the use of whichever "malloc"
is in effect, and thanks to the way reftable_malloc() is written, we
do not even have to be able to take the address of "malloc" ;-)

Will fast-track down to 'master'.

Thanks.
Junio C Hamano Jan. 8, 2025, 5:41 p.m. UTC | #3
René Scharfe <l.s.r@web.de> writes:

>> It is actually unnecessary to use those function pointers to
>> `malloc`/`realloc`/`free`, though: The `reftable` code goes out of its
>> way to fall back to the initial allocator when passing `NULL` parameters
>> instead. So let's do that instead of causing heap corruptions.
>
> Ugh.  That makes a lot of sense.  Sorry for the trouble! :-/

Thanks for a quick Ack.
diff mbox series

Patch

diff --git a/t/unit-tests/t-reftable-basics.c b/t/unit-tests/t-reftable-basics.c
index 990dc1a2445..1d640b280f9 100644
--- a/t/unit-tests/t-reftable-basics.c
+++ b/t/unit-tests/t-reftable-basics.c
@@ -157,13 +157,13 @@  int cmd_main(int argc UNUSED, const char *argv[] UNUSED)
 
 		old_alloc = alloc;
 		old_arr = arr;
-		reftable_set_alloc(malloc, realloc_stub, free);
+		reftable_set_alloc(NULL, realloc_stub, NULL);
 		check(REFTABLE_ALLOC_GROW(arr, old_alloc + 1, alloc));
 		check(arr == old_arr);
 		check_uint(alloc, ==, old_alloc);
 
 		old_alloc = alloc;
-		reftable_set_alloc(malloc, realloc, free);
+		reftable_set_alloc(NULL, NULL, NULL);
 		check(!REFTABLE_ALLOC_GROW(arr, old_alloc + 1, alloc));
 		check(arr != NULL);
 		check_uint(alloc, >, old_alloc);
@@ -188,11 +188,11 @@  int cmd_main(int argc UNUSED, const char *argv[] UNUSED)
 		arr[alloc - 1] = 42;
 
 		old_alloc = alloc;
-		reftable_set_alloc(malloc, realloc_stub, free);
+		reftable_set_alloc(NULL, realloc_stub, NULL);
 		REFTABLE_ALLOC_GROW_OR_NULL(arr, old_alloc + 1, alloc);
 		check(arr == NULL);
 		check_uint(alloc, ==, 0);
-		reftable_set_alloc(malloc, realloc, free);
+		reftable_set_alloc(NULL, NULL, NULL);
 
 		reftable_free(arr);
 	}