[Minor optimisation] Potential for optimisation in WriteToServer (SqlMetaDataPriv.IsNullable)

[thibault-reigner]

Hello,

While working on a project I have observed through DotTrace an opportunity to quickly get some better performance in WriteToServer :
Framework version : Net Framework 4.7.2
Microsoft.Data.SqlClient version : 2.1.2

As one can see, the HasFlag method used in SqlMetaDataPriv.IsNullable triggers some boxing and extra computation. I guess a solution like the one mentioned here https://medium.com/@arphox/the-boxing-overhead-of-the-enum-hasflag-method-c62a0841c25a
would result in some interesting performance gain as overall the method takes 0.41% of 9.18%, so ~4.5% of the WriteToServer method). This proportion could technically be higher since GetValueFromSourceRow calls the IDataReader supplied by the client code, whose performance could be better than the one from my example above and hence increase the relative proportion of ConvertValue and IsNullable.

[JRahnama]

@thibault-reigner thanks for sharing your idea to this repo. Can you make a PR for the suggested improvement otherwise we will get back to this in future when we have time.

[roji]

/cc @Wraith2 I think you were looking at this recently

[cheenamalhotra]

PR #1054

[Wraith2]

And #1064

From what I was told netcore 3.1 should generate masking for the HasFlag extension and that change was made in corefx when the code was only going to be used from netcore. I have seen this turn up on my own traces even under netcore so I'm not sure why it's not generating the same code as it would in sharplab. Manual flags is more compatible and doesn't box so everything will get switched over as soon as we can.