When running pstats with functions that take less than 0.5 millisecond (per call for example), the value shown is '0.000, which isn't really helpful.

This patch aims to show the value in microseconds instead of seconds for values that would otherwise be displayed as '0.000'

(the display may have to be tweaked for non-utf8 users, as the µ character is used)


Example output:

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
     9827    0.007      7µs    0.007      7µs {method 'get' of 'dict' objects}
     3427    0.009     25µs    0.017     49µs {map}
    10701    0.006      5µs    0.006      5µs {method 'lower' of 'str' objects}
    31410    0.045     14µs    0.045     14µs {method 'split' of 'str' objects}
    10023    0.006      6µs    0.006      6µs {_weakref.proxy}
    31801    0.009      2µs    0.009      2µs {len}
     3892    0.003      7µs    0.003      7µs {method 'extend' of 'list' objects}
     1397    0.001      8µs    0.001      8µs {method 'replace' of 'str' objects}
    10488    0.010      9µs    0.010      9µs {method 'string_literal' of '_mysql.connection' objects}
    10702    2.620   2447µs    2.620   2447µs {method 'readline' of 'file' objects}
    10023    0.004      3µs    0.004      3µs {method 'info' of '_mysql.connection' objects}
    10023    0.005      5µs    0.005      5µs {method 'insert_id' of '_mysql.connection' objects}
    14327    0.022     15µs    0.022     15µs {method 'rstrip' of 'str' objects}

Hum, "2447µs" makes the output more difficult to parse (by a program) and to read (by a human). Why not always displaying 6 digits (after the dot) for the two "percall" columns.

I figured we didn't want to change the size of the columns (+12 COL), but this could easily be done yes

I concur with Victor. This change likely breaks graphical viewers of pstat data.

How gprof and similar tools solve the problem of outputting sub-millisecond timings?

> I concur with Victor. This change likely breaks graphical viewers of pstat data.

I propose to *do* break the format by always displaying percall timings with a precision of 6 digits.

Based on the review comments, it looks like the existing PR 3931 needs to be revised and resubmitted against current master branch.  So, can we close PR 3931?