#7841 SchemaTool generates extra diff for platforms without FK support

[vpArth]

Add breaking test for #7841

[vpArth]

assertEmpty()

ok, changed

[vpArth]

Commits are squashed after both(scrutinizer, travis) checks passed

[lcobucci]

@vpArth thanks for sending this patch! Here are a few comments 👍

I also believe that this was intended as a bug fix, right? If so, it must use 2.6 as base instead of `master.

[vpArth]

it must use 2.6 as base instead of `master.

vendor/bin/phpcs fails every single file on 2.6 branch. is it ok?

[lcobucci]

it must use 2.6 as base instead of `master.

vendor/bin/phpcs fails every single file on 2.6 branch. is it ok?

@vpArth we have configured it to validate the diff in PRs because of that. You can focus on your changes only and ensure they're correct.

[vpArth]

Should I migrate to new PR to switch target branch to 2.6?

Upd: found option to switch the base =)

[lcobucci]

Awesome 🎉
Thanks!

[beberlei]

This is causing a regression in 2.6.5 and 2.7 for SQLite users: #7930

[beberlei]

This needs to be reverted. It fixes the symptom and not the problem.

1. SQLite does not support foreign keys, but we emulate them. A check for supportsForeignKeyConstraints() is the wrong place, because we interact with DBAL SchemaManager API in SchemaTool that abstracts the platform and if foriegn keys are emulated or supported.

2. Debugging into the code, the culprit is Comparator::diffTable() detects an "added foreign key", even though it already exists. That means that the SQLite abstraction does not populate Table#foreignKeys with the emulated keys.

3. The real fix is probably that we let SQLitePlatform#supportsForeignKeyConstraints() return true.

[beberlei]

Fix in DBAL: doctrine/dbal#3762

[lcobucci]

@vpArth we had to revert the changes as @beberlei explained. Unfortunately the fix can only be applied for ORM v2.8.0 since it requires us to bump up DBAL version (to 2.10).

[vpArth]

But what about #7841?
The SchemaTool bug for all platforms with false in supportsForeignKeyConstraints is still here then.

Adding true into SqlitePlatform could fix the regression without reverting that pr, right?

Have we any platform with correct populating of Table#foreignKeys with emulated keys?

[beberlei]

@vpArth please see doctrine/dbal#3762 - it ads the missing support in SQLite platform when it gets merged.