User-specified settings were being fiddled with by tidy internally. User
settings set by the user should always be able to be read back by the
user, but `AdjustConfig()` would change them. This change contributes
toward fixing the situation by using `AdjustConfig()` at the only point
that it's needed, as well by NOT automatically reverting to the snapshot
after outputting a buffer (which should be stateless, but caused PHP to
break because it tried to save buffer more than once, so subsequent calls
would use non-fiddle settings.).
Error made when merging LOADING Is. #879, PR #902
And merging SLOT PR #848
Also added a BIG warning over attribute_defs table to try to avoid this
in future.
* Is. #896 - make 'bear' docs match code
* Is. #487#462 add warn msg and do not get stuck until eof
The warning message could perhaps be better worded, and maybe there
should be another msg when a '>' is encountered while looking for a ']'
in a MS Word section, and perhaps the section should be discarded...
And perhaps it should be an error, to force the user to fix...
But the fix is good as it is, and these issues can be dealt with
later...
And this fix is piggy backed on this PR, but it is likewise related to
'word-2000' option...
* Support the <slot> tag
<slot> was added to the WHATWG HTML5 Living Standard on April 20, 2016.
* Also support the slot= attribute
Given that the <slot> tag by itself is not too useful, this commit also
introduces support for the global slot attribute.
* Using ParseBlock for <slot>
This is still suboptimal since ParseBlock will make it so that <slot>
always expects "Flow content", whereas the spec says that it should
have a Transparent content model.
In practice, it should allow all the cases that the spec allows for, but
it will also allow some cases that the spec does not allow. Notably, if
a <slot> tag is found in a Phrasing content (an inline context in
libtidy lingo), it will wrongly let Flow content (block tags in libtidy
lingo), whereas it shouldn't. But all in all, it's a good compromise.
This is a compromise -
1. Keep insertion of a blank title tag, if none.
2. Add new warn if title tag is blank, in html5
modified: include/tidyenum.h
modified: src/language_en.h
modified: src/message.c
modified: src/parser.c
The `setlocale` call doesn't return a single locale name in glibc when
any of the locale category variable has a different value, instead it
returns a composite locale name which is a concatenation of the entire
list of locale name and its values, causing the language detection fail.
This patch attempts to set the language via LC_MESSAGES and LANG
environment variables which are commonly used in POSIX-like systems,
then fallbacks to `setlocale` as the last resort.
