first shot remove TSRMLS_* things

Hash value must not zero?

fix the empty_strings SAPI shutdown leak in TS build

still the NTS variant is somewhat different as it needs
zend_new_interned_string_int because the normal callbacks might
be not in

fix the empty_strings SAPI shutdown leak in TS build

still the NTS variant is somewhat different as it needs
zend_new_interned_string_int because the normal callbacks might
be not initialized, but at least no leaks anymore and some more
structure

show more ...

Fixed useless or duplicated IS_INTERNED() checks

Changed "inline" into "zend_always_inline"

use zend_string_equals_ci

Add zend_string_equals and zend_string_equals_literal

master renames phase 4

master renames phase 2

master renames phase 1

first shot on merging the core fro the int64 branch

Fixed compilation warnings

Temporary fixed the invalid read warning

Changed zend_make_printable_zval() to return "use_copy" instead of additional reference argument.
Improved branch prediction.

C++ doesn't allow implicitly convert void * to other pointer type

C++ compiler doesn't allow cast a void * to other pointer type

Add STR_ALLOCA_* API, use in get_method and verify_arg

This avoid unnecessary allocations when using dynamic method
dispatch and class typehints.

Probably there are other places

Add STR_ALLOCA_* API, use in get_method and verify_arg

This avoid unnecessary allocations when using dynamic method
dispatch and class typehints.

Probably there are other places where this should be done as well,
those just stood out for my usage.

show more ...

Allocate zend_strings with correct size

For me (32bit) sizeof(zend_string) is 20, which means that the
char[1] array at the end is padded with three bytes. Thus allocating
based on s

Allocate zend_strings with correct size

For me (32bit) sizeof(zend_string) is 20, which means that the
char[1] array at the end is padded with three bytes. Thus allocating
based on sizeof(zend_string)-1 overallocates by those 3 padding bytes.

This commit fixes the allocation size, by using XtOffsetOf.

show more ...

Cleanup (2-nd round)

IS_INTERNED is not always 0 in ZTS

The current implementation chooses to intern "" in ZTS. Either that
should be removed or IS_INTERNED needs to actually check the flag.

Refactored ZVAL flags usage to simplify various checks (e.g. Z_REFCOUNTED(), candidate for GC, etc)

Changed data layout to allow more efficient operations

STR_DUP() doesn't duplicate interned strings anymore. In case new string is required STR_INIT() or STR_ALLOC() should be used.

Refactored GC (incomplete)

Don't include trailing zero into hash caclculation