array_reserve does not handle malloc errors

[alaviss]

Problem

The code allocates memory using malloc and realloc but does not check whether the allocation was successful.

For realloc, this meant the old allocation is leaked and the data is gone.
For malloc, this meant the array have a NULL contents pointer.

Either way, this is a SIGSEGV waiting to happen.

tree-sitter/lib/src/array.h

Lines 182 to 191 in 01bf431

 static inline void _array__reserve(Array *self, size_t element_size, uint32_t new_capacity) { 
   if (new_capacity > self->capacity) { 
     if (self->contents) { 
       self->contents = ts_realloc(self->contents, new_capacity * element_size); 
     } else { 
       self->contents = ts_malloc(new_capacity * element_size); 
     } 
     self->capacity = new_capacity; 
   } 
 } 

Steps to reproduce

N/A

Expected behavior

Checks are added to relay error to caller, or just abort the program on failure.

Tree-sitter version (tree-sitter --version)

tree-sitter 0.22.5

Operating system/version

Gentoo Linux

[Amaan Qureshi]

There is a check if the default ts allocator is used (which the cli does or if you specify -DTREE_SITTER_REUSE_ALLOCATOR) -

tree-sitter/lib/src/alloc.c

Line 5 in 8b28443

 static void *ts_malloc_default(size_t size) { 

If you are opting out of the ts allocator, then whatever the user supplies is used, or just the function itself if not.

[Max Brunsfeld]

Yeah, this working as intended. Like @amaanq, Tree-sitter aborts on allocation failure (unless you explicitly override that allocation behavior). Attempting to recover from allocation failure is explicitly not something we plan to do. For most codebases, attempting to do that correctly is a SEGV waiting to happen.

[alaviss]

Is it the intention that grammars should be complied with -DTREE_SITTER_REUSE_ALLOCATOR by default? None of the bindings seem to do that, so we already have SIGSEGV pending on OOM in all bindings.

At the very least I think the fallback implementations based on libc should also abort on failures.