WIP - template panel context issues

[kezabelle]

As I've now encountered an issue myself where the #910 changes caused problems, possibly related to some of the others, probably I should try and fix them all up.

As a short reminder, after template based widgets came along, and before #942 landed, I tried to ensure the number of calls to pprint.pformat was kept to a minimum, as it is an expensive calculation (#933).

In doing so, the call changed from: try: ... pformat(value); except: ... to use force_text instead of pformat. This was, in hindsight, a mistake, mostly around usages of Form instances.

Because __str__ on a Form will by default render the form, it has a number of side-effects:

• The repr won't get used if the rendering causes an SQL query to be triggered (ie: most forms!) (one I've now encountered)
• The form instance can have full_clean called on it because the __str__ includes error printing (one I've now encountered)
• Possibly subsequently prevents forms from rendering correctly at all? (1.8 seems to block form field rendering with Django 1.11. 1.7 works fine (though slow) #989)

---

In an ideal world, we could just change to repr (#725) instead of force_text, as this is similarly lightning fast. Instead, to err on the side of caution, I've opted for calling pprint.saferepr instead, which is protected against recursive data structures.
This has the benefit of still being faster than pformat (as far as I can tell) and avoids a certain class of problems the other options bring. But it's much slower than calling repr or force_text
directly.

---

Marking as WIP because there's a number of other issues (#989, #1024, #1053, #725) are related, and one option is just to undo the changes I made to avoid too many pformat calls. Also there's no tests ;)

[kezabelle]

Rough and ready benchmark of the different options:

from datetime import datetime
from pprint import pformat, saferepr
from django.utils.encoding import force_text

todump = {
    '1': 1,
    '2': '1',
    '3': None,
    '4': datetime.now(),
    '5': None,
    '6': object,
    '7': set(),
    '8': frozenset(),
    '9': [],
    '10': (),
}
if __name__ == '__main__':
    import timeit
    print(timeit.timeit('func(todump)', globals={'todump': todump, 'func': force_text}, number=50000))

swap out func for which-ever. Order, in lowest time to completion, should hopefully be:

• repr
• force_text
• saferepr (much slower than the 2 above)
• pformat (much slower than saferepr)

[codecov]

Codecov Report

Merging #1076 into master will not change coverage.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master    #1076   +/-   ##
=======================================
  Coverage   84.29%   84.29%           
=======================================
  Files          24       24           
  Lines        1318     1318           
  Branches      178      178           
=======================================
  Hits         1111     1111           
  Misses        157      157           
  Partials       50       50

Impacted Files	Coverage Δ
debug_toolbar/panels/templates/panel.py	85.93% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6647708...06bafad. Read the comment docs.

[matthiask]

Makes sense to me. Thanks!