On Thu, 10 Jan 2019, Riccardo (Jack) Lucchetti wrote:
On Thu, 10 Jan 2019, Sven Schreiber wrote:
> Am 10.01.2019 um 17:20 schrieb Riccardo (Jack) Lucchetti:
>> On Wed, 9 Jan 2019, Artur T. wrote:
>
>>>> Am 09.01.19 um 21:39 schrieb Alecos Papadopoulos:
>>>>> I have version 2018d Windows 64-bit.
>>>>>
>>>>> I try to generate random numbers using the "Random
variable..." command
>>>>> in the Add menu and I get
>>>>>
>>>>> <<'U1' -- no numeric conversion performed!>>
>
>>>
>>> I can confirm this using current git (2019a) on ubuntu linux.
>>
>> This should be now fixed in git.
>
> Good, I tend to think it's a pretty severe bug, for example for teaching
> purposes...
[I'm moving this to gretl-devel because it's fairly technical, but IMO wotrh
reading for people who like to see what's going on in the C sources]
In fact, it was pretty subtle: the C construct that was causing trouble was
na(x[j++])
and you would think: "OK, this evaluates if x[j] is missing, and
then moves on"; not quite. In fact, na(x) has become (after the big
NADBL reoganisation) a macro in which TWO calls are performed, so j
would go straight from 0 to 2. I performed a search for similar
construct in the whole source tree and fortunately there seem to be
no more instances.
Well figured out! Macros can be quite tricky.
Re. Sven's suggestion of a quick release with the fix: I'm not averse
to doing that. I've been working on our SVM support and the sooner we
release the sooner that support will be enabled in third-party builds
of gretl for Linux. (Up till now SVM has been optional, and dependent
on installed libsvm; now I've put a modified version of the libsvm
code into git so it's self-contained.)
Allin