Exception utils: remove redundant what() implementations

[tautschnig]

Make a std::string reason a member of the base class, and make
what() default to returning reason. This avoids duplicating the same
implementation across several child classes.

While at it, also fix the missing init-to-nil in one of
incorrect_goto_program_exceptiont's constructors.

• Each commit message has a non-empty body, explaining why the change was made.
• Methods or procedures I have added are documented, following the guidelines provided in CODING_STANDARD.md.
• n/a The feature or user visible behaviour I have added or modified has been documented in the User Guide in doc/cprover-manual/
• Regression or unit tests are included, or existing tests cover the modified code (in this case I have detailed which ones those are in the commit message).
• n/a My commit message includes data points confirming performance improvements (if claimed).
• My PR is restricted to a single feature or bugfix.
• n/a White-space or formatting changes outside the feature-related changed lines are in commits of their own.

[kroening]

The change means that there are two mechanisms for communicating the what message -- either via overloading what, or by setting reason. Pick one?

[codecov]

Codecov Report

Merging #6608 (86b13e9) into develop (afc53d8) will increase coverage by 0.00%.
The diff coverage is 84.00%.

@@           Coverage Diff            @@
##           develop    #6608   +/-   ##
========================================
  Coverage    76.73%   76.73%           
========================================
  Files         1579     1579           
  Lines       181999   181994    -5     
========================================
- Hits        139652   139650    -2     
+ Misses       42347    42344    -3     

Impacted Files	Coverage Δ
src/goto-programs/slice_global_inits.h	0.00% <0.00%> (ø)
src/goto-programs/string_instrumentation.h	0.00% <0.00%> (ø)
src/goto-programs/restrict_function_pointers.cpp	80.67% <100.00%> (+0.40%)	⬆️
src/memory-analyzer/gdb_api.h	100.00% <100.00%> (+9.52%)	⬆️
src/solvers/smt2/smt2_tokenizer.h	96.96% <100.00%> (ø)
src/util/exception_utils.cpp	97.77% <100.00%> (+3.43%)	⬆️
src/util/exception_utils.h	92.85% <100.00%> (-7.15%)	⬇️
... and 2 more

---

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 3f175c9...86b13e9. Read the comment docs.

[tautschnig]

The change means that there are two mechanisms for communicating the what message -- either via overloading what, or by setting reason. Pick one?

Isn't reason the raw data and what its formatted representation? I agree that it may be possible to rely on the default what() implementation by building the formatted representation in the constructor, but this also makes it impossible to later get back the raw data. I actually have this in #5837, where a get_reason() is added to one of the child classes.

[chrisr-diffblue]

Looks good, thanks!

[NlightNFotis]

LGTM, just one comment.

[NlightNFotis]

Is there a reason this was removed from the inheritance chain? Would it be better if message was removed and substituted with reason?

[tautschnig]

In this class, message is a std::ostringstream, making it really different from the parent class. Really, there is no meaningful commonality.

[peterschrammel]

The only benefit would be to catch all these exceptions at once, which could be done by another 'base class with reason' above the existing base, but that's minor IMO.