The native preg functions (preg_match, preg_replace, etc.) often require us to add a lot of useless boilerplate code to satisfy Phpstan, Scrutinizer, etc. Composer/Pcre offers us a way to remove that boilerplate, thereby giving us a cleaner codebase. I decided to try it on a few modules, and saw a result that clearly demonstrated the usefulness of doing this (aside from the cleaner codebase).
Sample 22_Reader_issue1767 reads an Xlsx spreadsheet with complex sheet names used in defined names, and writes it to Xlsx and Xls output files. When I changed Writer/Xls/Parser to use Composer/Pcre, this sample failed in many different places writing the Xls file. It turns out that some regexes were failing not because the string didn't match, but because the regex encountered "catastrophic backtracing". Composer/Pcre throws an exception when this happens; the native preg_match does return false, but we were not checking for that. The regexes in question are now changed to something which works, and formal unit tests are added for them. Finding this previously undetected error indicates that we should proceed with this change.
An alternative to using Composer/Pcre would be to test for false after all the preg calls. I have done this in the two samples changed with this PR. That seems adequate for a small number of changes, but it really just makes for more clutter considering the large number of regexps that we use in our code. I think Composer/Pcre is a better choice.
It isn't quite transparent. Composer forces all regexps to use PREG_UNMATCHED_AS_NULL, so some match fields will now be null instead of null-string (or non-existent if the unmatched field comes at the end). Our test suite doesn't report any problem (yet) due to this change, although Phpstan is sensitive to it. Several Phpstan annotations were eliminated due to this change, but some others are now needed.
It is not necessary to do this all at once. This PR addresses all the calls in Writer. I intend to address other components in several tickets.
The latest release seems to not want you to give a class element both a Php type and a doc-block type. I used the "fix" operand to delete the redundant doc-block declarations, with no other changes. So there should be no change to executable code.
While we might never be able to have 100% of our code strict, we can at
the very least do it for all of our tests. This ensures that our tests
are using our API with the types as intended by the test author, and not
silently be cast to what our API requires.
For one of the Phpstan upgrades, some message text had changed so drastically that the only practical solution at the time was to move the messages from phpstan-baseline.neon to phpstan.neon.dist. This was not ideal, but it allowed us time to move on and study the errors, which I have now done. At one point, Parser is expecting a variable to be an array, and that was not clear from the code. If not an array, the code will error out (which was Phpstan's concern); I have changed it to throw an exception instead. This satisfies Phpstan, and I can get the message out of neon.dist (without needing to restore it to baseline). Unsurprisingly, the exception was never thrown in the existing test suite, although I added a couple of tests to exercise that code.
In Helper/Dimension, Phpstan flagged a statement inappropriately. I suppressed the message using an annotation and filed a bug report https://github.com/phpstan/phpstan/issues/7563. A fix for the problem was merged yesterday, which is good, but it puts us in a tenuous position. The annotation is needed now, but, when the fix is inevitably pushed to the version we use, the no-longer-needed annotation will trigger a different message. Recode so that neither the current nor the future versions will issue a message, eliminating the annotation in the process.
oleibman
Adrien Crivelli