GitLab

A number of types, such as Dict views are not handled better, partially
due to the collections.abc change, and partially because these are now
visible in Python 3 where they were not in Python 2.

Collections.abc now collects a number of types - Callable and Iterable
in particular - which we use. These can be made to work with some more
conditional code.

In places the string version of the FrameFunction was giving us a
expanded version, where we might not expect it - adding the parameters
to the function name. Disabling this within the options is pretty
simple, and makes it possible to rely on the original behaviour
of the return value.

Classes not derived from object do not have an __class__ attribute
so we need to cope with this.

inspect.getmembers can fail on objects if they have a __get__ method
on members which raises an exception. This happens in the six module
when the lazy importing code is triggered, which causes a failure if
the import cannot work (as is the case with tkinter). We fall back
to just reporting that there are no members, which will cause the
presentation to default to its 'function name only', rather than the
full path naming, but will at least be safe.

It is now possible to cause class constants - values which are present
in an object which are the same as the class member of the same name -
to be hidden in the output. This can reduce the size of the output by
hiding information that can be obtained from the source itself.

Long lists and dictionaries are another place where the output can
become cluttered, and it is unlikely that the reader will want to check
the values beyond basic members of a dictionary, so these values can
be truncated as well - they would be truncated at the non-string limit
anyhow. Because the truncation happens at the item level, as well
as the total string level, individual items which are very long can
be kept from swamping the output. Additionally, the enumeration only
happens for sufficient items to meet the configuration, so the whole
dictionary or list does not have its representation generated only
to be truncated.

A small fix for the issues with the repr on StyledString not being
reported properly.

Check for exceptions in the shell code and cause an exit with a status
code of 1.

It seems that the old style setup.py includes wasn't working properly.

I'm trying to get the code to a point where it's suitable for uploading
to PyPI. Some of these files might be right; others might not, but right
now it's looking like it's got most of the right things in it.

When printing integers, it's useful to know what the hex value or whether
it is a specific bit. It won't hurt if you're not caring about those
values, but if you are, it's a lot easier than having to convert them
either in your head or starting up python to print them!

Lambda and Generator functions are handled better. Class methods are
detected slightly differently (but likely won't actually be any more
reliable - functions are returned for many results).

CodeNaming didn't have any tests, and could fail badly, which caused
the backtrace to report terminal errors. There is a test for one of the
functions in the CodeNaming which was giving problems - it should be
expanded as we find other issues, or when we need to report different
messages.

CI updated to the latest version, and updated the linting to use the
most recent pylintrc; with an annotation of an import that appears to
hide the top level longtraceback module but it'll be doing the right
thing anyhow.

The `experimental_repr` option enables the patching that gives more
information on the objects that we list. It's not perfect, but it's
got some useful features.

Errors whilst trying to process 'repr' will end up being shown as
an inline message (not through the error reporting system). This
means that the output might not match up quite so nicely with what
we hope to see, but it could help with some debugging.

There's no point in saying that this is an 'InteractiveConsole' as
the implementation isn't important. Better to say that it's Python
with the vesion number, which is closer to the standard version.

The 'imp' module is going away, to be reokaced by the 'importlib'
module. However, importlib, doesn't provide the 'new_module' support
that we require for starting the main environment. For simplicity,
this has been replaced with 'types.ModuleType', rather than going
through the Python 3.x interfaces of locating a spec (which seems
to not be possible for a null case).

This suffices for most cases; those cases of loading through a
loader like the Zip file, won't be handled, but then I've never
needed them.

In Python 2.7, the PWD was appearing as '' in the path, whilst in
Python 3.x, it appears as an absolute path. The former is not actually
the PWD, and would change in behaviour if the user changed directory.
The latter is more correct, and we reproduce this behaviour in the
shell for all versions of Python we're used with, so that we remain
consistent across versions. Even though that means that the shell
will differ from the regular python shell, I think that is more sane-
and anyhow, Python 2.7 is end of life so I think moving to the Python
3.x way of working is more sane.

ReprPatching isn't used at present, but if it were, it now handles the
regular expression pattern and match objects to give useful repr
output.

Small warning about a method possibly being a function.

Collecting artifacts only on failure isn't so helpful; we should pick
them up all the time.

If the frames filter list was empty, the checks for the end of the list
would generate an exception. A check added to shortcut this.

Attempt to get some junit results.

In python 3, the 'U' opening modifier is now deprecated. It's only
really necessary on Python 2 anyhow, so make the behaviour safe on
both (without warnings on Python 3).

Non-ascii strings were not being handled properly - although a lot
of the code was safe for handling the non-ascii string values, there
were some places which weren't handled properly - mostly in the
StyledString code. This has now been updated so that it works more
like we expect. It should now mean that we can report unicode strings
and UTF-8 strings properly in Python 2.7, and be sure to encode them
properly in Python 3+.

Better handling of Unicode/UTF-8 characters in Python 2.7.

The changed resolution order for files meant that the import didn't
work - now explicitly given a path.

Paste and others have special markers for the exclusion of frames from
the backtrace.

Styled string has been updated to have more tests and should be able
to report more problems. Unlikely that it'll cause any problems, but
it's good to keep it up to date.

When used within a threading environment, the race condition during the
creation of the singleton used in ReduceFilename causes aborts in the
constructor. This is awkward for us because it means that it is very easy
for us to end up with a crash if two tracebacks try to happen at the same
time. Fortunately, putting a lock around that code should make it safe
enough to use, I think.

When used within Python 3, the extract_* functions are actually returning
an object containing the elements, which returns as a tuple. In order
to support both the Python 2 and Python 3 usages, we use a namedtuple
to allow the interface to work whether indexed or by name.

The 'locals' parameter is currently ignored and always returns None.