Hi, I have a situation here where the "smpl ... --replace --restrict" construct isn't working like (I, at least) expected. Sorry, I can't give a minimal working example yet. The context is a panel dataset and we're inside a hansl function where the series y was passed as an argument, but series 'unit' was created locally inside the function from the $unit accessor. 'donor' and 'target_group' are valid integers that are actually present in the 'unit' series (I checked). smpl unit == donor --restrict matrix mydonor = {y} # no problem here, works smpl --full # should be redundant, but isn't -- perhaps a gretl bug? smpl unit == target_group --replace --restrict When I leave out the "smpl --full" line, I get an error message at the last quoted line saying "no observations would be left". Somehow it feels as if the "--replace" option here isn't honored. Because then the error message would make sense; since 'target_group' is a different unit than 'donor', putting the second restriction on top of the first would leave an empty set. Again, inserting the smpl --full line solves this, but it shouldn't be necessary, should it? This is with the latest snapshot. thanks sven _______________________________________________ Gretl-devel mailing list -- gretl-devel(a)gretlml.univpm.it To unsubscribe send an email to gretl-devel-leave(a)gretlml.univpm.it Website: https://gretlml.univpm.it/postorius/lists/gretl-devel.gretlml.univpm.it/

Am 18.01.2022 um 03:18 schrieb Cottrell, Allin: > I tried a simple example following your description and it worked > fine. We'll need to see what's the active ingredient in your case. Thanks, Allin. I was going to say my suspicion is that it's related to the recent panel-time changes, but then Artur reported he knows the problem already, in principle. [...]

Am 18.01.2022 um 17:58 schrieb Allin Cottrell: > Here's my simple case. Can you see what extra stuff you need to > introduce to provoke the problem? OK, thanks, this is helpful. I managed to provoke the same error message that I had before (no obs would be left): <hansl> function void pansmpl (series y, int val1, int val2)    series u = $unit    smpl u == val1 --restrict    print u    # workaround:    # smpl --full    smpl u == val2 --restrict --replace    print u end function nulldata 40 setobs 4 1:1 --stacked-time series y = normal() setobs 4 2000:1 --panel-time series mydates = $obsdate smpl mydates < 20001001    --restrict # discard last quarter pansmpl(y, 3, 7) </hansl>

Whether it's truly my problem I'm not 100% sure, but the setup looks quite similar indeed. Notice that the extra "smpl --full" line seems to solve it, like Artur also remarked.

OK, now we're in business. That's fixed in git.

One comment: using "mydates" and smpl --restrict is a kinda roundabout way of sampling panel data in the time dimension. You could do   smpl 2000:1 2000:3 --time or (with or without setting panel-time) just   smpl 1 3 --time

Am 18.01.2022 00:07 schrieb Sven Schreiber: > > > smpl unit == donor --restrict > matrix mydonor = {y}    # no problem here, works > smpl --full # should be redundant, but isn't -- perhaps a gretl bug? > smpl unit == target_group --replace --restrict > > When I leave out the "smpl --full" line, I get an error message at the > last quoted line saying "no observations would be left". > > Somehow it feels as if the "--replace" option here isn't honored. Hi, I've experienced this tricky issue for about 1,5 years, too. Allin had already fixed some corner case a while ago (discussed in a private chat). But there are still situations where the --replace option is not honored.

BTW: I surprised that the syntax "smpl --full" works at all as, according to the help text, it should be "smpl full".