Add INFECTION and TEST_TOKEN environment variables for each Mutant process

[maks-rafalko]

Add INFECTION and TEST_TOKEN environment variables for each Mutant process to make it possible to run functional tests for different databases.

Background:

We have a lot of functional tests that do real SQL queries to MySQL and decided to run tests in parallel. To achieve it, we integrated Paratest library.

Parallel processes can not write to the same database, because tests start to conflict with each other. To fix this issue, Paratest provides TEST_TOKEN=<int> environment variable for each process that can be used to set up different connections to the databases.

So if you have 3 parallel processes, these processes will use db_1, db_2, db_3 correspondingly.

Infection's issue with functional tests

The same goes for Infection. As soon as we start using more than 1 thread, our ParallelProcessRunner runs processes in parallel, and without TEST_TOKEN being passed, all those processes use the same database, which fails the test, making the majority of the Mutants killed.

This PR adds TEST_TOKEN=<int> env variable for each "mutant process".

Note: we can't use round-robin for thread indexes, because different processes take different amount of time. Consider the following example:

Imagine -we run Infection with 3 threads: infection -j3, and we have 4 mutations, so we need to run 4 processes when 3 of them can be in parallel

Process 1, TEST_TOKEN = 1
|--------------------------------------------------|
    Process 2, TEST_TOKEN = 2
    |------------|
         Process 3, TEST_TOKEN = 3
         |--|
             Process 4, TEST_TOKEN = 1
             |-----------------------------------------------------|

In this example, as soon as Process 3 finishes its work, we can run Process 4. It will mean that at the same time we will have exactly 3 parallel processes

• Process 1
• Process 2
• Process 4

But, since Process 1 takes too much time, round-robin algorithm gives TEST_TOKEN values of 1 for Process 4. And we have a situation where both processes Process 1 and Process 4 work with the same database at the same time.

To avoid such an issue, in this PR TEST_TOKEN gets the first available (free) index that is not used at this point of time by any other processes. For example above, as soon as Process 3 finishes its job, Process 4 will get TEST_TOKEN=3, because values 1 and 2 are still busy.

Process 1, TEST_TOKEN = 1
|--------------------------------------------------|
    Process 2, TEST_TOKEN = 2
    |------------|
         Process 3, TEST_TOKEN = 3
         |--|
-           Process 4, TEST_TOKEN = 1
+           Process 4, TEST_TOKEN = 3
             |-----------------------------------------------------|

INFECTION env variable

In our application, we have a custom bootstrup for PHPUnit, where we remove var/cache folder

# tests/bootstrap.php

$filesystem = new Filesystem();

$filesystem->remove([__DIR__ . '/../var/cache/test']);

Since Infection runs PHPUnit in separate processes, this bootstrap file gets called for each Mutant, again breaking the whole process, because while Process 1 can work, Process 3 removes its cache.

So, we need to understand when PHPUnit process is executed from Infection, and skip such removing:

$filesystem = new Filesystem();

if (!getenv('INFECTION')) {
    $filesystem->remove([__DIR__ . '/../var/cache/test']);
}

This PR:

• Covered by tests
• Doc PR: Add documentation for INFECTION and TEST_TOKEN environment variables site#214

[sanmai]

Is this something we can no longer test?

[maks-rafalko]

yes, since now we use range(1, $threadsCount), it leads to

1) Infection\Tests\Process\Runner\ParallelProcessRunnerTest::test_it_handles_all_kids_of_processes_with_infinite_threads with data set "infinite" (9223372036854775807)
ValueError: The supplied range exceeds the maximum array size: start=1 end=9223372036854775807

But I've added similar test in the next commit, where threads count is bigger than number of processes