8:39 p.m.
... AW: [Gretl-users] matrix rowname (without "s") similar
to colname; obs and functions
This thread is becoming quite complex. Maybe you could enter a short ticket
(bug or feature request, doesn't matter much) for this on the sourceforge
tracker.
Hi there, I did now find the time to go through my many questions and your many answers
again, to do as Sven suggested (quote) and opened a few tickets.
However, I have some more general remarks on my recent experiences with bundles (which are
great!), which I guess are better posted here.
Consider a list C in bundle B which is again located in bundle A. At some point, when
passing a bundle by reference to a function and then passing a part of that bundle by
reference to another function, a list that was part of the bundle became undefined (the
lest went away). I cannot reproduce this and perhaps I remember wrong. However, a bit of
the story I can reproduce:
<hansl>
nulldata 100
series A = normal()
list L = A
bundle boo
bundle coo
boo.coo = coo
eval inbundle(boo,coo)
boo.coo.L = L
eval inbundle(boo.coo,L)
function void testA(bundle Bndl)
eval inbundle(Bndl,L)
#eval argname(Bndl.L[1]) #error
eval varname(Bndl.L[1]) #OK
end function
function void testB(bundle *Bndl)
eval inbundle(Bndl,L) #not OK
end function
testA(boo.coo) #OK
catch testB(&boo.coo) #not OK
</hansl>
Including lists within bundles allows to do operations like sampling and manipulating
series located outside the bundle (which is great!), also varname(C[1]) would return the
variable name of the first element in list C. So basically this is a way to pass a series
by pointer, which is very useful. Perhaps this should be mentioned in the references?
Best
Frederik
9:36 p.m.
Am 22.01.2018 um 21:39 schrieb Schaff, Frederik:
Hi there, I did now find the time to go through my many questions
and
your many answers again, to do as Sven suggested (quote) and opened a
few tickets.
very good
Consider a list C in bundle B which is again located in bundle A.
This multiple nesting already sounds dangerous to me. Allowing lists in
bundles (which I pushed for I think) is a recent addition. It's good
that you're pushing the limits, though, to make things work that are
supposed to work.
catch testB(&boo.coo) #not OK
OT: catch is only supported on built-in functions, not user-defined ones
Including lists within bundles allows to do operations like sampling
and manipulating series located outside the bundle (which is great!),
I don't
have time right now to think about what was in your code and
what you meant, but this sounds dangerous as well; "manipulating
outside..." is not normally what we want to have/allow.
thanks,
sven
10:09 p.m.
> catch testB(&boo.coo) #not OK
OT: catch is only supported on built-in functions, not user-defined ones
> Including lists within bundles allows to do operations like sampling
> and manipulating series located outside the bundle (which is great!),
I don't have time right now to think about what was in your code and what
you meant, but this sounds dangerous as well; "manipulating outside..." is
not normally what we want to have/allow.
ah, the catch is not important. In my script I tested some more stuff afterwards and catch
allowed to move on and not stop on error (which is nice).
One way I use this frequently is a small function that adds noise to constant series, thus
allowing to draw a boxplot even if there is no variance. Another thing is to read in
variable names from outside series. Perhaps a "switch" to allow passing lists to
bundles is good, but please do not remove this option :). Btw., if you pass a list to a
function directly, this works the same (my noisyfied uses a list as input and creates for
each constant series a noisyfied substitute. Then it returns a list with original series
and noisyfied series, where appropriate)
Best
Frederik
9:50 a.m.
Am 22.01.2018 um 23:09 schrieb Schaff, Frederik:
> I don't have time right now to think about what was in your
code
> and what you meant, but this sounds dangerous as well;
> "manipulating outside..." is not normally what we want to
> have/allow.
but please do not remove this option :). Btw., if you pass a list to
a function directly, this works the same (my noisyfied uses a list as
Yes, it is a feature that when a function returns a list it can thereby
cause changes in the current workfile (manipulating series) that are not
explicitly specified. But it is important to understand that this is a
"lonely" exception to the rule that a function is encapsulated and will
normally only have effects that are explicitly given by the caller.
Having a list inside a bundle (and then perhaps nested in yet another
bundle...) sort of "hides" this exception even further from the caller
who may not expect such side effects from happening. This is what I
meant by "dangerous".
(And this kind of thing is also the reason why originally lists could
not be added to bundles, I believe.)
cheers,
sven
2533
days inactive
2534
days old
3 comments
2 participants
participants (2)
-
Schaff, Frederik -
Sven Schreiber