Skip to content

GitLab

Switch branch/tag
  1. 03 Jan, 2021 1 commit
  3. 02 Jan, 2021 4 commits
  5. 01 Jan, 2021 2 commits
  7. 31 Dec, 2020 2 commits
  9. 30 Dec, 2020 2 commits
  11. 28 Dec, 2020 1 commit
  13. 27 Dec, 2020 1 commit
  15. 25 Dec, 2020 1 commit
  17. 16 Dec, 2020 1 commit
    • Roman Perepelitsa's avatar
      fix(genpass): improve performance and usability and fix bugs (#9520) · b28665ae
      Roman Perepelitsa authored 
      *Bugs*

The following bugs have been fixed:

- All generators ignored errors from external commands. For example,
  if `/usr/share/dict/words` was unreadable, `genpass-xkcd` would
  print "0-" as a password and return success.
- All generators silently ignored the argument if it wasn't a number.
  For example, `genpass-apple -2` was generating one password and
  not printing any errors.
- All generators silently ignored extra arguments. For example,
  `genpass-apple -n 2` was generating one password and not printing
  any errors.
- `genpass-xkcd` was generating passwords with less than 128 bits of
  security margin in contradiction to documentation. The smaller the
  dictionary size, the weaker the passwords it was generating. For a
  dictionary with 27 words, `genpass-xkcd` was generating passwords
  with 93 bits of security margin (`log2(27!)`).
- The source of random data used by `genpass-xkcd` was not
  cryptographically secure in contradiction to documentation. See:
  https://www.gnu.org/software/coreutils/manual/html_node/Random-sources.html
- `genpass-apple` could generate a password with non-ascii characters
  depending on user locale. For example, passwords could contain 'İ'
  for users with Turkish locale.
- `genpass-apple` didn't work with `ksh_arrays` shell option.
- `genpass-xkcd` was printing spurious errors with `ksh_arrays` shell
  option.
- `genpass-xkcd` was producing too short (weak) or too strong (long)
  and/or printing errors when `IFS` was set to non-default value.
- All generators were printing fewer passwords than requested and
  returning success when passed a very large number as an argument.

*Usability*

Generators are now implemented as self-contained executable files.
They can be invoked from scripts with no additional setup.

Generators no longer depend on external commands. The only dependencies
are `/dev/urandom` and, for `genpass-xkcd`, `/usr/share/dict/words`.

All generators used to silently ignore all arguments after the first
and the first argument if it wasn't a number. For example, both
`genpass-apple -2` and `genpass-apple -n 2` were generating one password
and not printing any errors. Now these print an error and fail.

*Performance*

The time it takes to load the plugin has been greatly reduced. This
translates into faster zsh startup when the plugin is enabled.

Incidentally, two generators out of three have been sped up to a large
degree while one generator (`genpass-xkcd`) has gotten slower. This is
unlikely to matter one way or another unless generating a very large
number of passwords. In the latter case `genpass-xkcd` is now also
faster than it used to be.

The following table shows benchmark results from Linux x86-64 on i9-7900X.
The numbers in the second and third columns show how many times a given
command could be executed per second. Higher numbers are better.

command                     | before (Hz) | after (Hz) | speedup |
----------------------------|------------:|-----------:|--------:|
`source genpass.plugin.zsh` |        4810 |      68700 |  +1326% |
`genpass-apple`             |        30.3 |        893 |  +2846% |
`genpass-monkey`            |         203 |       5290 |  +2504% |
`genpass-xkcd`              |        34.4 |       14.5 |    -58% |
`genpass-xkcd 1000`         |       0.145 |      0.804 |   +454% |
      b28665ae
  19. 14 Dec, 2020 4 commits
  21. 13 Dec, 2020 2 commits
  23. 12 Dec, 2020 3 commits
  25. 11 Dec, 2020 1 commit
  27. 10 Dec, 2020 3 commits
  29. 09 Dec, 2020 4 commits
  31. 08 Dec, 2020 1 commit
  33. 07 Dec, 2020 6 commits
  35. 05 Dec, 2020 1 commit