Theses newly added macros will be used to implement the equivalent under
MSVC. Also, move the definitions to headers, where they make more sense,
and for some, are even more useful there (e.g. malloc).
- Use the extensions autoconf finds for object and executable files.
- Remove the sorev variable, and replace SOREV definition with sorev's.
- Default to je_ prefix on win32.
Using errno on win32 doesn't quite work, because the value set in a shared
library can't be read from e.g. an executable calling the function setting
errno.
At the same time, since buferror always uses errno/GetLastError, don't pass
it.
MSVC doesn't support C99, and building as C++ to be able to use them is
dangerous, as C++ and C99 are incompatible.
Introduce a VARIABLE_ARRAY macro that either uses VLA when supported,
or alloca() otherwise. Note that using alloca() inside loops doesn't
quite work like VLAs, thus the use of VARIABLE_ARRAY there is discouraged.
It might be worth investigating ways to check whether VARIABLE_ARRAY is
used in such context at runtime in debug builds and bail out if that
happens.
MSVC doesn't support C99, and as such doesn't support designated
initialization of structs and unions. As there is never a mix of
indexed and named nodes, it is pretty straightforward to use a
different type for each.
Fix a potential deadlock that could occur during interval- and
growth-triggered heap profile dumps.
Fix an off-by-one heap profile statistics bug that could be observed in
interval- and growth-triggered heap profiles.
Fix heap profile dump filename sequence numbers (regression during
conversion to malloc_snprintf()).
Since we're now including jemalloc_internal.h, all the required headers
are already pulled. This will avoid having to fiddle with headers that can
or can't be used with MSVC. Also, now that we use malloc_printf, we can use
util.h's definition of assert instead of assert.h's.
Remove mmap_unaligned, which was used to heuristically decide whether to
optimistically call mmap() in such a way that could reduce the total
number of system calls. If I remember correctly, the intention of
mmap_unaligned was to avoid always executing the slow path in the
presence of ASLR. However, that reasoning seems to have been based on a
flawed understanding of how ASLR actually works. Although ASLR
apparently causes mmap() to ignore address requests, it does not cause
total placement randomness, so there is a reasonable expectation that
iterative mmap() calls will start returning chunk-aligned mappings once
the first chunk has been properly aligned.
Fix chunk_alloc_dss() to zero memory when requested.
Fix chunk_dealloc() to avoid chunk_dealloc_mmap() for dss-allocated
memory.
Fix huge_palloc() to always junk fill when requested.
Improve chunk_recycle() to report that memory is zeroed as a side effect
of pages_purge().
Fix a memory corruption bug in chunk_alloc_dss() that was due to
claiming newly allocated memory is zeroed.
Reverse order of preference between mmap() and sbrk() to prefer mmap().
Clean up management of 'zero' parameter in chunk_alloc*().
Using static memory when malloc_tsd_malloc fails means all threads share
the same wrapper and thus the same wrapped value. This defeats the purpose
of TSD.
Not setting the initialized member leads to randomly calling the cleanup
function in cases it shouldn't be called (and isn't called in other
implementations).
These flags take unsigned values, but they were fed with signed values
taken with va_arg, and that led to sign extension in cases where the
corresponding value has the most significant bit set.
This allows for different patterns for file names:
- lib.so.version for e.g. Linux
- lib.version.dylib for OSX (which is much more common than
lib.dylib.version)
- lib.dll for Windows (no version at all).
Clean up a few config-related conditionals to avoid unnecessary
dependencies on prof symbols. Use cassert() rather than assert()
everywhere that it's appropriate.
Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
Change the "opt.prof_accum" default from true to false.
Add the "opt.prof_final" mallctl, so that "opt.prof_prefix" need not be
abused to disable final profile dumping.
Mike Hommey