By parsing php://input in chunks, we can bypass max_input_vars for
forms which do not use the "multipart/form-data" enctype. This can
be (and is) safely skipped if there are fewer than max_input_vars
fields submitted as part of $_POST.
This adds a category expanded which:
* fetches child content in a category tree if it has not already been loaded;
* toggles relevant classes on the category node to show and hide child content; and
* applies appropriate animations to improve user experience.
For example, a user may create a date/time profile field and set the 'Start year'
field to '2012'. Another user, using the Hijri calendar, may edit the name of
this field. They visit the settings page and in this case the date '1/1/2012'
is converted to '7/2/1433' in Hijri. So, the year '1433' is then displayed.
The user then changes the name of this field and saves the form. This is where
the issue occurs, as the date '1/1/1433' is converted into Gregorian, which
converts to the year '26/09/2011', so the year '2011' is saved in the DB, not
'2012'.
The usergetdate function is not only used to display dates, it is also passed
to the function make_timestamp. This means users using another calendar type
other than Gregorian will generate incorrect timestamps which may be saved in
the DB or used to populate the date_selector and date_time_selector elements.
For example, when creating an assignment using another calendar type other
than Gregorian, the mod_form.php file calls $this->apply_admin_defaults()
which uses the function usergetmidnight to set the "Allow submissions from"
date to today's date and the "Due date" field to 7 days in the future. The
usergetmidnight function calls usergetdate which is then passed to
make_timestamp. Since the usergetdate function was using the calendar
type's timestamp_to_date_array function the date being passed to
make_timestamp was not in Gregorian. So, when using the hijri calendar the
year 1434 was being passed which was generating a large negative number as
the timestamp which was then used to populate the date fields.
Renamed the userdate/usergetdate functions in the calendar type to be more
descriptive and made them abstract to ensure developers implement this
functionality in their calendar type.
Also tidied up PHPDocs for these functions.
1) Moved the calendar types location to a sub-folder in the calendar directory.
2) Removed/moved language strings.
3) Removed calendar types that should be downloaded as plugins.
4) Removed a Non-English language pack for the Gregorian calendar type that
should be downloaded separately.
5) Removed custom files responsible for checking for updates and uninstalling
calendar types, which should be done by core.
6) Removed JS from the calendar_month block as there is no non-JS alternative
provided and the JS written does not make use of the YUI library to ensure
multiple browser support.
7) Removed code from the base class responsible for creating timestamps that
are saved in the DB.
8) Added PHPDocs.
Note: In the original patch we are editing core functions responsible for saving
time in the database in the calendar base class. This is very dangerous, we do
not want to touch these functions as it could cause a complete fubar of the
database. There are places we are forcing the use of the gregorian calendar as
we are passing dates generated by the PHP date function, where as sometimes
we pass dates from usergetdate (which was being overwritten to return the
date specific to the calendar). We can not expect third party modules to
change the calendar type depending on the format they pass to these functions.
Filter on enrolled users page use role param to perform filtering by role,
and unassign role also contain role param to remove user role in course
which used to conflict. Looking at assign role, we use roleid param, so
using roleid for unassigning as well
While working on the issue, I spotted these two places that were worth of
fixing. The first one is a trivial reminiscence of a previous refactoring,
after which both branches of the if() statement became equal.
The second one is actually a typo as in theory it could generate unexpected
input fields with the name like qPP1. Luckily this never happened due to the
way how survey questions are hardcoded (there are no questions with the type 2
that would require two answers to their subquestions).
In MDL-7501, we stopped using rowspanning cells in the form table due to
accessibility. That had introduced a regression so in the COLLES P+A survey,
all the rows were displayed with the same background colour. This patch returns
the previous behaviour that each couple of items can be distinguished by the
background colour.
Also, there is no need to display "I prefer that" and "I found that" as a small
text any more. It had made sense in rowspanning layout but not after MDL-7501
was fixed.
And finally, as all items are enumerated now sequentially, there are actually
48 lines, each couple covering one question in two variants. I think it's
correct to reflect this in the description of the form so the text was slightly
amended.
This basically reverts the commit 5196df589b that
I believe was a result of misunderstanding of how question type field is
(ab)used in the Survey module. As it took significant time to get familiar with
the overall logic of questions and their processing in the module, I left my
findings in added inline comments. The point is that it is $question->type that
matters. Types of questions listed as subquestions in the multi field is
irrelevant in that case (and all have it set to 1 IIRC).
This patch re-enables the "COLLES (Preferred and Actual)" survey type that did
not work at all due to regression.
