Config: only determine screen width if/when needed (performance improvement)

[jrfnl]

Description

Recreation of upstream PR squizlabs/PHP_CodeSniffer#3831:

Config: only determine screen width if/when needed

I'm not sure about *nix, but on Windows, the call to shell_exec('stty ...') is slow.

As things were, the call to shell_exec('stty ...') would be:

• Made when the initial default value for reportWidth is being set in restoreDefaults().
• Potentially made a second time if the users CodeSniffer.conf file contained a report_width => 'auto' entry.
• Potentially made a third time if any of the rulesets used for the run contained a <config name="report-width" value="auto"/> entry.
• Potentially made a fourth time if --report-width=auto would be passed from the command-line.

This is inefficient for the following reasons:

1. The output from shell_exec('stty ...') does not change between calls (well, providing the user doesn't resize their terminal in the microseconds between calls)
2. We don't actually need to know the value 'auto' translates to, until the reportWidth will be used.

Taking the above into account, making the same call up to four times is not desirable.

This commit moves the translation from 'auto' to an actual terminal width from the __set() method to the __get() method and overwrites the reportWidth value from 'auto' with the actual terminal width value, if available, and with the default value if the terminal width could not be determined.
This means that subsequent calls to __get() for the reportWidth will automatically use the previously determined value instead of trying to determine value again.

This removes the inefficiency and should make PHPCS runs a little bit faster (at the very least on Windows).

The only time multiple calls to shell_exec('stty...') could now need to be made, would be if the reportWidth would be changed (back to 'auto') between the first retrieval and a subsequent retrieval of the reportWidth value. As things are, this will never happen in a normal PHPCS run, though could possibly happen in a test situation.

AbstractMethodUnitTest: take advantage of the change in reportWidth handling

For the tests using the AbstractMethodUnitTest class, the reportWidth and most other config settings are irrelevant.

This commit changes some of the set up/tear down for the test to make the use of the Config class more efficient.

This should make the tests using the AbstractMethodUnitTest class as their base significantly faster (at the very least on Windows).

While not benchmarked properly, I have done some comparisons with the test runs on my local machine on Windows.

• phpunit --filter Core (= the tests which use this base class + a few extra tests):
  Before: 2 minutes.
  After: 8 seconds.
• The same effect can be seen when running phpunit without a --filter:
  Before: 7 minutes.
  After: 5 minutes.
• And when I apply a similar change to the one made here to the base test class in PHPCSUtils (4621 tests):
  Before: 2.5 minutes.
  After: 1 second.

Suggested changelog entry

• Runtime improvement for PHPCS CLI users. The improvement should be most noticeable for users on Windows.

Related issues/external references

Follow up on squizlabs/PHP_CodeSniffer#3761 for which tests were added in squizlabs/PHP_CodeSniffer#3820.