This needs to go into 5.4 as well, but will wait for Pierre to review win32 situation
# Patch by Lucas Nealan, Arnaud Le Blanc, Brian Shire & Ilia Alshanetsky
directive is set in activation time). This commit fixes this by adding a per
request parsing of the browscap file that's when get_browser is called the
first time and the directive is set on activation time.w
This reverts the fix for bug #24557.
- Make php_stream_free delete the stream from the resources list, not merely
decrease its refcount, as a single call to zend_list_delete does.
#Not worth the risk merging to 5.3. While change #2 may prevent some segfaults,
#a quick and dirty survey to the codebase only showed calls to php_stream_close
#or php_stream_free on streams allocated in the same function, which would have
#refcount == 1. May be reconsidered.
wrapper).
#php_stream->wrapperdata should hold an array zval (like its zval* type
#indicates...), it's not a place where the wrapper can drop an arbitrary
#pointer. For that, .wrapperthis should be used.
#Also, since the ftp dir wrapper defines its own stream type, it's more
#appropriate to use .abstract to store the stream instance specific data.
setting).
#This fix (for g/G/k/H modes) is done at a different level than that for the
#modes e/E/f/F, at a bit higher level and therefore with less coverage. I
#chose this because it addresses the problem where it is -- the calling function
#that passes a buffer too small to php_gcvt.
stream when 0 is given as the value.
- PHP_STREAM_OPTION_WRITE_BUFFER no longer changes the chunk size in socket
streams.
- Added stream_set_chunk_size() function.
- Some signedness fixes.
- Test for commit r308474, now that it's possible to actually test it.
negative ports and ports > 65535 as invalid.
The tests that fail due to #308035 in the standard ext were not
fixed. If the behavior in those tests turns out to be the
desirable one, both this commit and #308035 ought to be reverted
or at least adapted.
to be called as all the headers are being sent and after all
of the default headers have been merged.
headers_list(), header_remove() and header() can all be used
inside the callback.
<?php
header('Content-Type: text/plain');
header('X-Test: foo');
function foo() {
foreach (headers_list() as $header) {
if (strpos($header, 'X-Powered') !== false) {
header_remove('X-Powered-By');
}
header_remove('X-Test');
}
}
$result = header_register_callback('foo');
echo "a";
- Fixed usage of zend_multibyte_get_internal_encoding (its return cannot be
cast to char*).
- Change tests to reflect that charset detection now relies on
internal_encoding, not on current_internal_encoding.
NOTE: This fixes the changes in rev 306077, but it remains that that change
introduced a BC break. I assumed it was intentional
- Added mechanism to force outer streams to be closed before their inner ones.
- Fixed temp:// streams only handling correctly (through an ad hoc mechanism) reverse closing order
when the inner stream is of type memory.
being UTF-8.
- Moved ext/standard/tests/file/htmlentities19.phpt and htmlspecialchars.phpt to the
proper directory (tests/strings instead of tests/file). htmlentities19.phpt had to
be renamed in order not to clash with an existing file.
effect)
#Made "from" a proper INI setting and bound it to a global variable.
#Previously, it was simply read from the hash table with the parsed ini file
#by using cfg_get_string (I wonder why this mechanism still exists...)
of reported malformed sequences). (Gustavo)
#Made a public interface for get_next_char/utf-8 in trunk to use in utf8_decode.
#In PHP 5.3, trunk's get_next_char was copied to xml.c because 5.3's
#get_next_char is different and is not prepared to recover appropriately from
#errors.
remains.
- Fixed bug on determine_charset that was preventing correct detection in
combination with internal mbstring encoding "none", "pass" or "auto".
- Added profiles for entity encode/decode for HTMl 4.01, XHTML 1.0, XML 1.0
and HTML 5. Added the constants ENT_HTML401, ENT_XML1, ENT_XHTML and
ENT_HTML5.
- htmlentities()/htmlspecialchars(), when told not to double encode, verify
the correctness of the existenting entities more thoroughly.
It is checked whether the numerical entity represents a valid unicode code
point (number is between 0 and 0x10FFFF). If using the flag ENT_DISALLOWED,
it is also checked whether that numerical entity is valid in selected
document. In HTML 4.01, all the numerical entities that represent a Unicode
code point (< U+10FFFFFF) are valid, but that's not the case with other
document types. If the entity is not valid, & is encoded to &.
For named entities, the check is also more thorough. While before the only
check would be to determine if the entity was constituted by alphanumeric
characters, now it is checked whether that entity is necessarily defined for
the target document type. Otherwise, & is encoded to &.
- For html_entity_decode(), only valid numerical and named entities (as defined
above for htmlentities()/htmlspecialchars() + !double_encode) are decoded.
But there is in this case one additional check. Entities that represent
non-SGML or otherwise invalid characters are not decoded. Note that, in
HTML5, U+000D is a valid literal character, but the entity 
 is not
valid and is therefore not decoded.
- The hash tables lazily created for decoding in html_entity_decode() that were
added recently were substituted by static hash tables. Instead of 1 hash
table per encoding, there's only one hash table per document type defined in
terms of unicode code points. This means that for charsets other than UTF-8
and ISO-8859-1, a conversion to unicode code points is necessary before
decoding.
- On the encoding side, the ad hoc ranges of entities of the translation
tables, which mapped (in general) non-unicode code points to HTML entities
were replaced by three-stage tables for HTML 4 and HTML 5. This mapping
tables are defined only in terms of unicode code points, so a conversion
is necessary for charsets other than UTF-8 and ISO-8859-1. Even so, the
multi-stage table is much faster than the previous method, by a factor
of 5; the conversion to unicode is a small penalty because it's just a
simple table lookup.
XML 1.0/htmlspecialchars() uses a simple table instead of a three-stage
table.
- Added the flag ENT_SUBSTITUTE, which makes htmlentities()/htmlspecialchars()
replace the invalid multibyte sequences with U+FFFD (UTF-8) or &#FFFD;
(other encodings).
- Added the flag ENT_DISALLOWED. Implements FR #52860. Characters that cannot
appear literally are replaced by U+FFFD (UTF-8) or &#FFFD; (otherwise).
An alternative implementation would be to encode those characters into
numerical entities, but that would only work in HTML 4.01 due to limitations
on the values of numerical entities in other document types. See also the
effects on htmlentities()/htmlspecialchars() with !double_encode above.
fix the erratic behavior without breaking backwards compatibility. Namely,
$offset retains SEEK_SET behavior but actually SEEK_CUR is passed to
_php_stream_seek, if possible, by moving the offset stream->position bytes.
- Addresses bug #53006.
current position (seek with SEEK_CUR, not SEEK_SET). Only positive values are
allowed. This breaking change is necessary to fix the erratic behavior in
streams without a seek handlder. Addresses bug #53006.
#Note that the example on the doc page for stream_get_contents() may fail
#without this change.
#This change is also in the spirit of stream_get_contents(), whose description
#is "Reads all remaining bytes (or up to maxlen bytes) from a stream...".
#Previous behavior allowed setting the file pointer to positions before the
#current one, so they wouldn't be "remaining bytes". The previous behavior was
#also inconsistent in that it allowed an moving to offset 1, 2, ..., but not 0.
* Fixed typo: "Unknow" -> "Unknown"
* Removed useless Win9x version info
# I will merge this to 5.3 once I have analyzed a possible bug
# (and hopefully fixed) why Server 2008 is reported as unknown
hint, like htmlentities et al.
- Fixed bug #49407 (get_html_translation_table doesn't handle UTF-8).
- Fixed bug #25927 (get_html_translation_table calls the ' ' instead of
').
- Fixed tests for get_html_translation_table and unified the Windows and
non-Windows versions of the tests.
code unit sequences to integers so as to store the entity
maps. Code in traverse_for_entities assumed little
endian. Fixed.
(in practice, due to the absence of unicode and entity
mappings for multi-byte encodings -- except UTF-8 --, this
doesn't matter, so the relevant code was commented out for
performance reasons).
- (5.3) Fixed bug in the Windows implementation of
dns_get_record, where the two last parameters wouldn't be
filled unless the type were DNS_ANY (Gustavo).
"raw mode". In this mide, $type (2nd parameter) is the numeric type of the record, and
the responses are not parsed -- the "type" element will be numeric and there will be
a "data" element with the raw data of the response buffer, which the programmer will
have to parse.
- Fixed bug in the Win32 implementation of dns_get_record, where the 3rd and 4th arguments
would only be filled if the 2nd ($type) was DNS_ANY.
- [DOC] The 3rd and 4th parameters can now be NULL (changed their arginfo).
- Dramatic improvements on the performance of html_entity_decode and htmlspecialchars_decode, as the
string is now traversed only once. Speedups of 20 to 25 times with Windows release builds and a
~250 characters string (for 2nd and subsequent calls).
- Consistent behavior on html_entity_decode. For instance, the entity in "&<" would be decoded,
but not "&é". Not anymore. The code path for "basic" and non-basic entities is now mostly
shared.
- Code of html_entity_decode and htmlspecialchars_decode is now shared.
- [DOC] More consistent behavior of htmlspecialchars_decode. Instead of translating only <, >,
&, ", ' and ', now e.g. ", ', ', ', etc. are also decoded.
- [DOC] Previous translation of unicode code points in numerical entities was seriously broken. When
the code points for some character were not the same in unicode and the target encoding, the
behavior could be an erroneous translation (e.g. 0x80-0xA0 in win-1252) or no translation at all.
Added unicode translation tables for all single-byte encodings. Entities are not translated for
multi-byte entities, except for ASCII characters whose code points are shared. We could add
the huge translation tables (several thousand elements) for those encodings in the future.
- Fixed numerical entities that after # had text accepted by strcol being accepted.
- Much more commented and well-structured code...
- Tests for get_html_translation_table()) are broken. I stared fixing the tests, but then I realized
it was completely helpless because get_html_translation_table() is broken by not handling
multi-byte characters correctly.
* Deprecated getrandmax() and mt_getrandmax() in favour of the new constants
# We should promote constants for static data like such instead of function calls
# maybe this was just an old left-over
