Fixed some issues with running the test suite which would be revealed by
running tests in random order.

Fixes: #338
Change-Id: I90c2a86f9279a265f64b75be3a7af314ab8772ba

Fixed Python deprecation issues related to module importing, as well as
file access within the Lingua plugin, for deprecated APIs that began to
emit warnings under Python 3.10.  Pull request courtesy Petr Viktorin.

This modernizes the code to avoid `DeprecationWarning` and `ResourceWarning` encountered in the test suite under Python 3.10a4:
- [load_module](https://docs.python.org/3/library/importlib.html#importlib.abc.Loader.load_module) is deprecated
- Some files weren't being closed

This changes the semantics of the `compat.load_module` function: on Python 3.5+, the module is no longer inserted in `sys.modules`. All non-test calls did `del sys.modules[self.module_id]` right after the call, anyway. On older Python, the module is inserted and then deleted.

(Some additional `DeprecationWarning` come from Setuptools: https://github.com/pypa/setuptools/pull/2517)

Closes: #328
Pull-request: https://github.com/sqlalchemy/mako/pull/328
Pull-request-sha: 87c1d09e5f39de7b19e749568c437928580f7553

Change-Id: I9cb3772f6812ef14297792344caf2f9aa5337adf

The default template encoding is now utf-8.  Previously, the encoding was
"ascii", which was standard throughout Python 2.   This allows that
"magic encoding comment" for utf-8 templates is no longer required.

Fixes: #267
Change-Id: I411a75ebd6161e2517e06b5180a3571e5f5d923e

While I can believe that the standard exception might not have worked
with some ancient version of pytest, it certainly works just fine now.
This saves people from importing nose which kills pytest
via DeprecationWarnings these days.

Closes: #316
Pull-request: https://github.com/sqlalchemy/mako/pull/316
Pull-request-sha: 90dfcfc3d0351ab9bf8125e8c2adb45c57059a6d

Change-Id: Ie816acd325c6783e08801bc83de37dbe6b4581e5

Replaced usage of the long-superseded "parser.suite" module in the
mako.util package for parsing the python magic encoding comment with the
"ast.parse" function introduced many years ago in Python 2.5, as
"parser.suite" is emitting deprecation warnings in Python 3.9.

Additionally repair two error-detection unit tests that were using a
"print" statement despite the tests running under Python 3; while the
tests were passing they were no longer testing the right thing.

Fixes: #310
Change-Id: Id436e2c6620438ef16b3f188b989233c82972b29

Mako 1.1 now supports Python versions:

* 2.7
* 3.4 and higher

This includes that setup.py no longer includes any conditionals, allowing
for a pure Python wheel build, however this is not necessarily part of the
Pypi release process as of yet.  The test suite also raises for Python
deprecation warnings.

Fixes: #249

Replaced usage of ``inspect.getfullargspec()`` with the vendored version
used by SQLAlchemy, Alembic to avoid future deprecation warnings.  Also
cleans up an additional version of the same function that's apparently
been floating around for some time.

Fixes: #295
Change-Id: I98274c16b6022289d1890f4daf532bab323ab112

In some situations, it is inconvenient to pass default_filters in the
Template constructor depending on the template in question.  It might be
easier in such situations to express page filters in the template itself.
However, dropping the existing default_filters (either explicitly set or the
default of ["str"] resp. ["unicode"]) might break existing templates.

The code change here comes to the rescue in such situations.  Existing
templates keep working as they are, but editors of templates get a tool to
replace the default filters for specific templates.  They do take on the
responsibility of turning all encountered inputs into strings, lest they
fail along the lines of https://github.com/sqlalchemy/mako/issues/272.

This change should be sufficiently backwards compatible to not cause any
concerns.  Sure, technically a "n" filter at page tag level was treated as a
no-op so far.  So theoretically existing templates could break.  But there
was no incentive to have such an "n" filter at the page tag level, and the
expressed semantics of the "n" filter is to suppress default filters, so
semantically anyone relying on it being a no-op in that situation was using
unsupported hacks anyway.

Closes: #299
Pull-request: https://github.com/sqlalchemy/mako/pull/299
Pull-request-sha: f8d5a22db3230634d2b42c59909985f31875a9f5

Change-Id: Ide030975229c1df7c0cef534976f740a03c17ca6

Using URIs as entrered will make the stack trace arguably more readable than
using the module name which is the URI with non-word characters replaced by
underscores.

This change also fixes a mistake where a mods cache hit would not update
template_filename, causing a stale filename to be displayed if the stack
trace alternates between different templates.  There now is a test case for
this.

Closes: #298
Pull-request: https://github.com/sqlalchemy/mako/pull/298
Pull-request-sha: ea35208d504932561711cdb574ec1f7def1e7060

Change-Id: Ieb606f6b6b7f4602d4d56694dd0bccf6dc287d20

Improved the line-number tracking for source lines inside of Python  ``<%
... %>`` blocks, such that text- and HTML-formatted exception traces such
as that of  :func:`.html_error_template` now report the correct source line
inside the block, rather than the first line of the block itself.
Exceptions in ``<%! ... %>`` blocks which get raised while loading the
module are still not reported correctly, as these are handled before the
Mako code is generated.  Pull request courtesy Martin von Gagern.

Closes: #297
Pull-request: https://github.com/sqlalchemy/mako/pull/297
Pull-request-sha: 60ad749604f8de0a9b8133430995045a93d0939c

Change-Id: I2086e4370e8e2db7099de01743840286f0160efc