Set length variable for slice and index expressions during semantic

[ibuclaw]

Optimization opportunity noticed in #7677.

For static arrays, slice and indexes using values known at compile time should now never generate array bounds checks at codegen.

Before:

int[10] sa;
enum n = 2;                  // upperInRange
enum m = 5;                  // | lowerLessThan
                             // | |  check code
{ auto s = sa[0 .. 0];     } // 1 1  NULL
{ auto s = sa[0 .. $];     } // 0 1  ($ <= 10           )
{ auto s = sa[1 .. $];     } // 0 0  ($ <= 10 && 1 <= $ )
{ auto s = sa[$ .. $];     } // 0 0  ($ <= 10 && $ <= $ )
{ auto s = sa[0 .. $/n];   } // 0 1  ($/n <= 10         )
{ auto s = sa[$/n .. $];   } // 0 0  ($ <= 10 && 5 <= $ )
{ auto s = sa[$/m .. $/n]; } // 0 0  ($/n <= 10 && $/n <= $/m)

After

int[10] sa;
enum n = 2;                  // upperInRange
enum m = 5;                  // | lowerLessThan
                             // | |  check code
{ auto s = sa[0 .. 0];     } // 1 1  NULL
{ auto s = sa[0 .. $];     } // 1 1  NULL
{ auto s = sa[1 .. $];     } // 1 1  NULL
{ auto s = sa[$ .. $];     } // 1 1  NULL
{ auto s = sa[0 .. $/n];   } // 1 1  NULL
{ auto s = sa[$/n .. $];   } // 1 1  NULL
{ auto s = sa[$/m .. $/n]; } // 1 1  NULL

For reviewing the test changes (mostly just whitespace alignment with the rest of the function).
https://github.com/dlang/dmd/pull/7682/files?w=1

[dlang-bot]

Thanks for your pull request, @ibuclaw!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

[ibuclaw]

@nordlow - we can do this at least. :-)

[nordlow]

@ibuclaw - great

[wilzbach]

closed and reopened to restart SemaphoreCI (after moving it to my account, s.t. @ibuclaw can login there via SSH too). If you want to be able to do the same, login once via GitHub at Semaphore and then let me know.

[wilzbach]

Shouldn't there be tests in fail_compilation for cases where constant folded detected an error?

[wilzbach]

Apparently this caching doesn't work.

[ibuclaw]

You mean coverage wise? That's probably because semantic only calls it once. Doesn't make the safeguards any less correct.

[wilzbach]

Looks like this doesn't work.

[ibuclaw]

Control flow has likely moved this to the other function. Should still be kept around though, as it's a utility function.

[wilzbach]

l <= 10 ?

[wilzbach]

1 - not l ?

[ibuclaw]

I should probably change my font, because all I see is the same character twice in your comment. I assume you mean there's a mixup between number (1) and letter (L)?

[wilzbach]

Yes, the non-monospace font shows it more clearly:

• { auto s = da[l .. u]; } // 0 0 (u <= 10 && l <= u )
• { auto s = da[1 .. u]; } // 0 0 (u <= 10 && l <= u )

[JinShil]

I was trying to figure out why that was even compiling. Line 190 shows size_t l = 0; // upperInRange, so it looks ok, though it could probably use a better name.

[wilzbach]

1 - not l ?

[wilzbach]

(rebased with trivial conflicts)

[ibuclaw]

Rebased, clarified names of vars in the bounds tests.

[ibuclaw]

These two can be improved in a latter PR by using the check for 0 or $ from #7677.

[JinShil]

Closed and reopened in an effort to get Jenkins to pick up the latest bits.

[ibuclaw]

Really need to fix that...

[wilzbach]

Hopefully dlang/ci#163 helps a bit.

[ibuclaw]

Rebased, again.