[Sebb767]

> "I think this is stupid" is a really poor reason to break backwards compatibility, despite how many other software projects use this reasoning.

True. But his argument seems to be in the direction of "this is highly unexpected behaviour" and I tend to agree. The number of applications broken by extending the date range probably dwarves to the number of bugs avoided by not having the time span break at such a strange length.

[falcolas]

> "this is highly unexpected behaviour"

The behavior in question is having an upper and lower bound on a time interval. This strikes me as highly expected behavior.

The maximum on the interval is lower than the author expected; and reading the docs quickly cleared up what the interval maximum is.

The entire rant boils down to "MySQL's choice to keep backwards compatibility is stupid, because I think this interval limit should be larger."

[deschutes]

The unforgivable thing here isn't the anachronism -- it's the lack of a suitable replacement. If you want a full i64 worth of duration at a reasonable resolution, you're on your own.

[simias]

Reading the article it seems like the core issue is:

>To make matters worse, it appears that the most popular EF Core MySQL provider maps .NET’s TimeSpan to TIME by default, despite the fact that TimeSpan can contain intervals in the dozens of millennia (it uses a 64 bit integer and has 10-8 s precision)

It's this mapping is bogus. If the ranges don't match TimeSpans should be mapped to something else.

MySQL's BIGINTs are 64bits, so that might be a more reasonable target. Of course you lose some "typing" information in the schema, but it's probably worth it.

But I suppose it would make sense for MySQL to provide a "BIGTIME" alias for this purpose, and deprecate the old one maybe.

[zaphar]

On the one hand no reaches for MySQL because it has robust types you can use for the columns. On the other hand not having a robust useful Interval type does not speak well of MySQL in this day and age.

[falcolas]

It's a type with a minimum and maximum. How in the world is this an "unforgivable" sin? Shall I rail on C for the `size_t` type only having 16 bits?

> lack of a suitable replacement

MySQL's various string "alternatives" have proven how stupid having multiple types with different limits really can be.

[Sebb767]

I agree it's not a necessary change. But, on the other hand, how much compatibility would it really break? I cannot imagine much applications are dependent on MySQL throwing an error at 800-odd hours (and it would arguably be a very strange design decision).

[erik_seaberg]

MySQL is notorious for quietly guessing what you must have meant. Writers need to validate up front, and then try to read back what was actually written.

[toast0]

Having a type mean different things on different versions of the database is not good. If you want a new type that does new things, it should have a new name. If the old type is so bad, you could make an sql mode to blacklist it.

Yes, that leaves a legacy of why should I use TIME2 or whatever instead of TIME, etc. That's the downfall of having a successful project over a long time frame and not having designed it perfectly at the beginning.

See also UTF8 shouldn't be used in MySQL because it's dumb. But you can't change UTF8 to do the right thing, you have to have a new name that's better.

[Dylan16807]

It would be a huge change in behavior to fix utf8.

Changing the limit from 838 to 999 wouldn't.

[falcolas]

> how much compatibility would it really break

Probably only a few applications; but that's still too many.

And backwards compatibility does matter. A change here means that some few dozen developers have to now troubleshoot a previously-stable application which now fails silently in odd corner cases.

[nytgop77]

then create new type. time2 or datetime13 with improved behaviour and say in docs to prefer newer unless your specicaly know that you need old behaviour.