Add feature to allow personalised error messages per field.

[ispyinternet]

I've added tests and updated the readme if you are happy to include this feature.

[coveralls]

Coverage remained the same at 100.0% when pulling 5f2f7a7 on ispyinternet:master into f83d71b on icebob:master.

[icebob]

Thank @ispyinternet, great PR, but this solution causes performance degradation. I think there is no reason to manage custom messages in validation time because it can be managed in compilation time and performance won't be affected. What do you think? Could you modify the PR?

[ispyinternet]

@icebob I've tried to adapt. Please review.

[icebob]

Could you explain what is this code?

[ispyinternet]

In order to remove check from run time we have to prepare all possible message types at compile time. These include:

‘required’
‘string’ | ‘number’ | ‘array’ etc (i.e type)

Then we analyse the additional rules to see what other messages need to be added. When analysing the rule object, obviously we ignore the other properties that we know are not additional rules, i.e. 'type', 'items' for arrays, 'props' for objects and our new 'messages' property.

[ispyinternet]

I've just changed the if statement to use key in obj style rather than truthy test

[icebob]

Good job, the performance has been improved! On my PC 5.3M ops/sec. The master branch is 5.2M 👍

[icebob]

please don't force push commit because i'm trying to linting the codes.

[ispyinternet]

sorry, I wont anymore

[icebob]

Great! So is it done? Can I merge?

[ispyinternet]

All tests are passing, not sure if you want to review performance or do any further sanity checks. Some unexpected changes to review.

[icebob]

The compile is very bad. It comes with the latest commit. I think due to Object.keys because it is a very expensive function.
This PR:
√ compile & validate 28,840 rps
√ validate with pre-compiled schema 5,359,590 rps
√ validate with wrong obj 668,274 rps

Master:
√ compile & validate 130,815 rps
√ validate with pre-compiled schema 5,123,820 rps
√ validate with wrong obj 657,710 rps

[ispyinternet]

Also for every rule we are now iterating over all the default messages. Previously only iterates if user supplied. Perhaps an overwriting merge of the two objects. I’ll make the update. Is there a command to run so I can get the metrics to report back?

[icebob]

I've added a condition, so only handle custom messages if there is messages property in schema.

[icebob]

Now it's much better.
It doesn't affect the performance without messages, and with it now is ~70k

√ compile & validate 119,438 rps
√ compile & validate with custom messages 69,858 rps
√ validate with pre-compiled schema 5,220,102 rps
√ validate with wrong obj 660,502 rps

By the way, you can check perf with npm run bench

[ispyinternet]

Nice! Why did you remove Object.keys(this.messages) from constructor? Now it is called everytime a message property is encountered - negligible admittedly!?

Also did you see the slight changes to lib/rules/string.js and lib/rules/email.js?

[icebob]

Because there is not relevant difference. With Object.keys ~68k, with static array ~70k. But I'm checking again.

In string & email files I made some linting in regexp.

[icebob]

@ispyinternet I put back the key array. It gives +10%. So now I think it's ready & performant. Could you add this new feature to the README too with examples?

[ispyinternet]

I already updated the README its the last section

[icebob]

Really, sorry :)

[icebob]

Thank you for your work.

[ispyinternet]

Thanks for your help too!

[ispyinternet]

you can close #28