make_bytes() assumed that if the Content-Encoding header is set, then
everything had already been dealt with bytes-wise, but in a streaming
situation this was not necessarily the case.
make_bytes() is only called when necessary when working with a
StreamingHttpResponse iterable, but by that point the middleware has
added the Content-Encoding header and thus make_bytes() tried to call
bytes(value) (and dies). If it had been a normal HttpResponse,
make_bytes() would have been called when the content was set, well
before the middleware set the Content-Encoding header.
This commit removes the special casing when Content-Encoding is set,
allowing unicode strings to be encoded during the iteration before they
are e.g. gzipped. This behaviour was added a long time ago for #4969 and
it doesn't appear to be necessary any more, as everything is correctly
made into bytes at the appropriate places.
Two new tests, to show that supplying non-ASCII characters to a
StreamingHttpResponse works fine normally, and when passed through the
GZip middleware (the latter dies without the change to make_bytes()).
Removes the test with a nonsense Content-Encoding and Unicode input - if
this were to happen, it can still be encoded as bytes fine.
Backport of 250aa7c39b from master.
It prevented the GZipMiddleware from compressing some data types even on
more recent version of IE where the corresponding bug was fixed.
Thanks Aaron Cannon for the report and Tim Graham for the review.
This (nearly) completes the work to isolate all the test modules from
each other. This is now more important as importing models from another
module will case PendingDeprecationWarnings if those modules are not in
INSTALLED_APPS. The only remaining obvious dependencies are:
- d.c.auth depends on d.c.admin (because of the is_admin flag to some
views), but this is not so important and d.c.admin is in
always_installed_apps
- test_client_regress depends on test_client. Eventually these should
become a single module, as the split serves no useful purpose.
For users who didn't activate autocommit in their database options, this
is backwards-incompatible in "non-managed" aka "auto" transaction state.
This state now uses database-level autocommit instead of ORM-level
autocommit.
Also removed the uses_autocommit feature which lost its purpose.
enter_transaction_management() was nearly always followed by managed().
In three places it wasn't, but they will all be refactored eventually.
The "forced" keyword argument avoids introducing behavior changes until
then.
This is mostly backwards-compatible, except, of course, for managed
itself. There's a minor difference in _enter_transaction_management:
the top self.transaction_state now contains the new 'managed' state
rather than the previous one. Django doesn't access
self.transaction_state in _enter_transaction_management.
There were a couple of errors in ._dirty flag handling:
* It started as None, but was never reset to None.
* The _dirty flag was sometimes used to indicate if the connection
was inside transaction management, but this was not done
consistently. This also meant the flag had three separate values.
* The None value had a special meaning, causing for example inability
to commit() on new connection unless enter/leave tx management was
done.
* The _dirty was tracking "connection in transaction" state, but only
in managed transactions.
* Some tests never reset the transaction state of the used connection.
* And some additional less important changes.
This commit has some potential for regressions, but as the above list
shows, the current situation isn't perfect either.
