Use the faster table insertion method (dlg_settings_advanced.lua only)

[ghost]

No description provided.

[sofar]

Not really critical or anything hotpath, but still fine I think.

[ghost]

Not really critical or anything hotpath

Yes. But even not critical, I think the faster the better.

[kilbith]

On non-critical code you need readability and ease of maintenance more than a micro-optimization.

There are other ways to make the insertion faster, but it would be dumb to lose the readability for 3 ms. faster here...

[paramat]

👍

[kaeza]

tinsert(t,v)      2365273
t[#t+1]=v         2522097
n=n+1 t[n]=v      7909498

The numbers are insertions per second using different approaches (tested under LuaJit, stock Lua may be different; benchmark here if you want to test).

Still agree with @kilbith.

[sofar]

lua-5.1.2, tested with the pasted link by kaeza:

DESC              TOT
tinsert(t,v)      1938457.3333333
t[#t+1]=v         2280419.3333333
n=n+1 t[n]=v      3821620.6666667

[sofar]

For reference, here is a full spread of test data across versions:

Numbers are insertions per second

=== lua-5.1.4 ====
DESC              TOT
tinsert(t,v)      1894375.3333333
t[#t+1]=v         2231591.6666667
n=n+1 t[n]=v      3757611
=== lua-5.2.3 ====
DESC              TOT
tinsert(t,v)      1820607.6666667
t[#t+1]=v         2076934.3333333
n=n+1 t[n]=v      3404882.3333333
=== lua-5.3.2 ====
DESC              TOT
tinsert(t,v)      1957116.6666667
t[#t+1]=v         2088806.6666667
n=n+1 t[n]=v      3481470.6666667
=== luajit ====
DESC              TOT
tinsert(t,v)      2392461
t[#t+1]=v         2597967
n=n+1 t[n]=v      4398963.3333333

My biased opinion:

• It's nonsense to convert from tinsert() to t[#t + 1], it's maybe 5% difference
• Incrementing a table index is way, way faster (50%)

[ghost]

I will close this PR and revert 24e8b0a with new PR.