Warning on PERSISTENT statement

I know that PERSISTENT statement is slow, and try to overcome by passing an logical argument to by-pass when I know the function doesn't need using the persistent variable. This illustrate by my FOO function

function a = foo(astatic)

if astatic
    % mlint: "PERSISTENT could be very inefficient unless it is a top-level
    % statement in its function"
    persistent A
    a = A;
else
    a = 3;
end

end % of foo

But then MATLAB gives the warning I quote abobe.

To clear thing out, I made a comparison of runing time with BAR function

function a = bar(astatic)
% No warning
persistent A

if astatic
    a = A;
else
    a = 3;
end

end % of bar

And then when I test (R2019a, windows) I get those results

tic;
for k=1:1000000
    foo(0); 
end
toc % Elapsed time is 0.076238 seconds.

%%
tic;
for k=1:1000000
    foo(1); 
end
toc % Elapsed time is 0.784971 seconds.

%%
tic;
for k=1:1000000
    bar(0); 
end
toc % Elapsed time is 0.764904 seconds.

%%
tic;
for k=1:1000000
    bar(1); 
end
toc % Elapsed time is 0.790119 seconds.

As you can see, the results do not backup MLINT message.

Can I just ignore it or is there anything else I miss?

[Adam Danz]

Probably the same story as 'global'

https://www.mathworks.com/matlabcentral/answers/19316-global-could-be-very-inefficient#answer_25760

[Walter Roberson]

I had forgotten that question, but amazingly I think my logic then still holds up.

Oh, and the resolution of conflicts between local and global has changed since that was posted, so my lack of clarity on the point was wise.

[Bruno Luong]

I don't think PERSISTENT would be implement as GLOBAL.

Runing time, persistent vs global: I can't see why the MATLAB somewhere else where the PERSISTENT variables exists. It is right there just like any local variable, however the internal mxArrary structure should be stored somesort of heap/static memory zone and it must not be cleared as local variables when the function exits.So PERSISTENT does not require MATLAB to b search in some sort of data base (using hash?), and it should be faster than GLOBAL. And it is (about 7/8 fold) as showed here:

function a = glob(astatic)

if astatic
    global A
    a = A;
else
    a = 3;
end

end % of glob

function a = foo(astatic)

if astatic
    % mlint: PERSISTENT could be very inefficient unless  it is a top-level
    % statement in its function 
    persistent A
    a = A;
else
    a = 3;
end

end % of foo

%%
tic;
for k=1:1000000
    glob(0); 
end
toc % Elapsed time is 0.099111 seconds.
%%
tic;
for k=1:1000000
    glob(1); 
end
toc % Elapsed time is 5.843688 seconds.

%%
tic;
for k=1:1000000
    foo(0); 
end
toc % Elapsed time is 0.076238 seconds.

%%
tic;
for k=1:1000000
    foo(1); 
end
toc % Elapsed time is 0.784971 seconds.

%%
tic;
for k=1:1000000
    bar(0); 
end
toc % Elapsed time is 0.764904 seconds.

%%
tic;
for k=1:1000000
    bar(1); 
end
toc % Elapsed time is 0.790119 seconds.

My my question is indeed similar to other post about the correctness of MLINK message and it is obviously a wrong warning for both global and persistent usage.

[Rik]

So actually the warning should be "this usage can lead to unexpected rersults"?