PR: Add permanent history to the debugger

[impact27]

Description of Changes

Small changes to pdb to save the history, so I can access the history of a previous debugging session.

This is not a really nice way of doing thing, but it works. Two (non easy) way of improving this:

• The history of pdb should be saved at the spyder-kernels level, as is the case with ipython.
  (Well only kind of because the history is duplicated, see PR: Load history from IPython directly #7141 )
• The history should be saved in a sqlite file, in the same way ipython does. I tried to reuse ipython core functions but I could't get it to work quickly. This allows to save the time the command was executed, and to separate between different sessions.

• Wrote at least one-line docstrings (for any new functions)
• Added unit test(s) covering the changes (if testable)

• Included a screenshot or animation (if affecting the UI, see Licecap)

Issue(s) Resolved

Fixes #

Affirmation

By submitting this Pull Request or typing my (user)name below,
I affirm the Developer Certificate of Origin
with respect to all commits and content included in this PR,
and understand I am releasing the same under Spyder's MIT (Expat) license.

I certify the above statement is true and correct:

[pep8speaks]

Hello @impact27! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file spyder/plugins/ipythonconsole/widgets/debugging.py:

Line 231:80: E501 line too long (80 > 79 characters)

Comment last updated at 2019-08-23 13:52:17 UTC

[ccordoba12]

@impact27, I left a couple of minor comments, otherwise looks good.

[ccordoba12]

This is not a really nice way of doing thing, but it works

I know this is not the best solution, but it's good enough until we have time to seriously improve our debugger (which is not going to happen for Spyder 4, but probably for version 5).

[impact27]

The only think that could be

This is not a really nice way of doing thing, but it works

I know this is not the best solution, but it's good enough until we have time to seriously improve our debugger (which is not going to happen for Spyder 4, but probably for version 5).

I have been playing with that a bit and there is a problem with multilines inputs. I will try to fix that with sqlite.

[ccordoba12]

I have been playing with that a bit and there is a problem with multilines inputs. I will try to fix that with sqlite.

You shouldn't waste your time with that because the solution we're thinking to implement doesn't involve enhancing IPdb but creating a debugging kernel.

Besides, I think multiline is really not an option in Pdb because you have to prefix Python code with ! so that Pdb parses it correctly.

[impact27]

I used from IPython.core.history import HistoryManager so the history will be correctly saved.

[ccordoba12]

I used from IPython.core.history import HistoryManager so the history will be correctly saved.

I'm -1 on that and I'd like you to revert that change. That not only makes things more complex than they need to be, but also there's no need of it because you can write multi-line statements in Pdb.

[impact27]

Well the complexity is all in IPython.core.history.HistoryManager, so the spyder part is actually simpler. The multiline problem comes when I have:

import numpy as np
np.abs([1,
        2])

If I want to copy-paste the abs in pdb, it will work as is:

But in the old implementation, as a newline separate the statement, I ended up with:

history = [
'import numpy as np', 
'np.abs([1,',
'        2])']

With the new implementation, the newlines are not specific characters, so there is no problem.

Using IPython.core.history.HistoryManager adds the benefits that sessions are correctly handled. It is possible to separate the history of two different session, event if they took place at the same place.

Futhermore, all the tools that are used to manage ipython history can be easily adapted to manage this new pdb history.

I therefore believe that this implementation is more future proof and has less bugs.

[impact27]

I mean with a database I can store executed cells, not just lines, and I can attach more informations, such as the session this line was from, the line number, and more. So in the future if we want to add functionalities to the history, the database already exists and we don't need to migrate from a text file.

[ccordoba12]

The multiline problem comes when I have

I haven't considered this use case (I don't know how it could arise, besides copy/paste).

I mean with a database I can store executed cells, not just lines

Cells executed with runcell, right?

So in the future if we want to add functionalities to the history, the database already exists and we don't need to migrate from a text file.

Ok, that's a good point too.

[ccordoba12]

I left a couple of minor comments, but I'd like to mention two other things:

1. I think the main changes could be moved to DebuggingWidget, right? I mean, I don't see anything in them that requires the ControlWidget interface now.

2. A test for this is missing. It can consist on creating a console, entering debug mode and writing a command on it, then opening a new console and verifying that that command was saved correctly in the debugging history.

   That test can replace the contents of current test_browse_history_dbg test, present in test_ipythonconsole.py.

[impact27]

Cells executed with runcell, right?

Multiline statment pasted by the user in pdb, but it still needs to be a single statment. otherwise:
*** SyntaxError: multiple statements found while compiling a single statement

[ccordoba12]

@impact27, this now needs a merge with master.

[impact27]

There is a small issue with backward search in pdb that I will have a look at now.

[ccordoba12]

Thanks @impact27!! Another great addition of yours!