Reverts #268 and properly fix what was failing

[HavokInspiration]

#268 was a mistake.
This should properly fix it.

#268 introduced a regression when baking a diff : you ended up with all your schema being in the diff where existing tables / columns had a changeColumn registered for no reason.

[HavokInspiration]

Oh, and I also removed 5.4 from the tests matrix.

[HavokInspiration]

On second thoughts, this could still be a problem the other way around.
The root cause of the problem is from the core : the per-column collation was added. Which means that when a table is reflected to its SchemaTable object counter-part, each column now has a new collation column.

So if you had a dump from before 3.2.12 (when per column collation was added), your dump did not contain the collation column. If you were to update to 3.2.12 and bake a diff, you'd experience the problem described in the original post of the PR.
If you had a dump from after (with the collation column) but used migrations 1.6.4, you had the problem I described above (that's how I encoutered it : CakePHP 3.3.9 and migrations 1.6.4 and my full schema was in the diff file each time).

So it should be known that if people with a dump that does not have the collation column now bake a diff, they will end up with all their schema being in the diff where columns from existing tables will have a changeColumn call attached.

[HavokInspiration]

Or we could remove collation from both the schema from the dump and from the database and we would not have a problem anymore whatever the versions of Cake / plugin used.

Since phinx does not support per-column collation (last time I checked), this is actually not a problem to remove it.

Any opinion on this ?

[HavokInspiration]

I pushed a new commit where I play defensive and unset the collate option from both the schema read from the table and the schema from the dump. It seems safer this way and prevents undesired side-effects.
And since per-column collation is not supported right now, this should not be such a big-deals.

This PR was reported in issue #283 as well.

[codecov-io]

Current coverage is 84.15% (diff: 100%)

Merging #279 into master will increase coverage by <.01%

@@             master       #279   diff @@
==========================================
  Files            31         31          
  Lines          1671       1672     +1   
  Methods         213        213          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits           1406       1407     +1   
  Misses          265        265          
  Partials          0          0          

Powered by Codecov. Last update a7d36f1...5228366

[lorenzo]

Hmm too bad phone does not support collate. This fix makes sense to me

[HavokInspiration]

@lorenzo The only real collate support I have seen in phinx is with MySQL on a per-table basis.