R help - Jan 2022 - Error: if statement: missing value where TRUE/FALSE needed

I've created a git repository
at?https://sourceforge.net/p/rhelp/code/ci/master/tree/.? I've used git
before, so that's what I chose.? I had forgotten that sourceforge didn't
have a download all files command.? I don't post files often.

About the semantics of "crash":? Perhaps I'm showing my age here.?
It used to be that programs crashed, bombed, etc. when they hit unanticipated
errors.? These programs lived in a OS without an intermediary.? Now, we have
programs of programs running within multiple applications on multiple servers
with multiple OSs.? Since I'm not a programmer by profession, I've had
little need to use these constructs.

Chuck Coleman



On Wednesday, January 5, 2022, 02:51:07 PM EST, Duncan Murdoch
<murdoch.duncan at gmail.com> wrote:





On 05/01/2022 2:09 p.m., Ivan Krylov wrote:
> On Wed, 5 Jan 2022 21:22:37 +0300
> Ivan Krylov <krylov.r00t at gmail.com> wrote:
> 
>> How exactly do you check for a missing argument?
> 
> To answer my own question, the check is
> 
>>> browser(expr = {
>>>? is.na(w1) | is.na(ub) | is.na(m) | is.na(wi) | is.na(lb)
>>> })
> 
> I think that there are NAs in xseg, which I don't see a check for. In
> particular, a few lines above, there's
> 
>>> md = median( xseg, na.rm = TRUE);
>>> if (robust) sdd = mad(xseg,na.rm = TRUE)
>>> else sdd = sd(xseg,na.rm = TRUE)
> 
> ...which seems to suggest they are expected to happen there.
> 
> Not sure which advice to give regarding the choice of file hosting
> service. On the one hand, SourceForge lets you upload a whole directory
> of files in arbitrary formats, but isn't very convenient do download
> more than one file from. On the other hand, "pastebin" services
like
> http://paste.debian.net/ or http://paste.org.ru/ let us browse the code
> right away, without having to look at the "Your download will start
> shortly..." screen, but don't host *.RData or more than one file
per
> link. I guess that some other code hosting service like
> SourceHut/GitHub/GitLab/Gitea/... could fulfil both criteria.

I would take one step back, and suggest that if any project has as many 
scripts as this one, it should be put in an R package.? You can 
distribute that by building it into a tar.gz file and putting that 
single file in a place where people can download it.? Github and R-forge 
work well for packages, and I assume other hosting services do too, 
though I haven't used many.

People seem to be unreasonably reluctant to put their code into packages.

Duncan Murdoch

I've downloaded your code, and I'm seeing some strange results.  When I 
source MetroTest.R, it creates a log file ending

Error in if (xseg[w1] > ub & m == 1) olrmark[wi] = TRUE else if 
(xseg[w1] <  :
   missing value where TRUE/FALSE needed

as you saw.  But when I comment out the sink() lines in it, I don't see 
an error.  If I comment out only the first sink line, i.e. run with

   my_log <- file("MetroTest.log")
   #sink(my_log, append = TRUE, type = "output") # Writing console 
output to log file
   sink(my_log, append = TRUE, type = "message")

I get an error on the last line:

 > sink(my_log, append = TRUE, type = "message")
Error in sink(my_log, append = TRUE, type = "message") :
   'file' must be NULL or an already open connection

Once running this code R crashed.

I find all of that quite perplexing.  I suspect one of the packages has 
some buggy code, but I don't know which one.  Currently I'm running the 
script after running gctorture(); that's really slow.  If it turns up 
anything I'll let you know.

Duncan Murdoch



On 06/01/2022 10:23 a.m., Chuck Coleman wrote:
> I've created a git repository
at?https://sourceforge.net/p/rhelp/code/ci/master/tree/.? I've used git
before, so that's what I chose.? I had forgotten that sourceforge didn't
have a download all files command.? I don't post files often.
> 
> About the semantics of "crash":? Perhaps I'm showing my age
here.? It used to be that programs crashed, bombed, etc. when they hit
unanticipated errors.? These programs lived in a OS without an intermediary.?
Now, we have programs of programs running within multiple applications on
multiple servers with multiple OSs.? Since I'm not a programmer by
profession, I've had little need to use these constructs.
> 
> Chuck Coleman
> 
> 
> 
> On Wednesday, January 5, 2022, 02:51:07 PM EST, Duncan Murdoch
<murdoch.duncan at gmail.com> wrote:
> 
> 
> 
> 
> 
> On 05/01/2022 2:09 p.m., Ivan Krylov wrote:
> 
>> On Wed, 5 Jan 2022 21:22:37 +0300
>> Ivan Krylov <krylov.r00t at gmail.com> wrote:
>>
>>> How exactly do you check for a missing argument?
>>
>> To answer my own question, the check is
>>
>>>> browser(expr = {
>>>>  ? is.na(w1) | is.na(ub) | is.na(m) | is.na(wi) | is.na(lb)
>>>> })
>>
>> I think that there are NAs in xseg, which I don't see a check for.
In
>> particular, a few lines above, there's
>>
>>>> md = median( xseg, na.rm = TRUE);
>>>> if (robust) sdd = mad(xseg,na.rm = TRUE)
>>>> else sdd = sd(xseg,na.rm = TRUE)
>>
>> ...which seems to suggest they are expected to happen there.
>>
>> Not sure which advice to give regarding the choice of file hosting
>> service. On the one hand, SourceForge lets you upload a whole directory
>> of files in arbitrary formats, but isn't very convenient do
download
>> more than one file from. On the other hand, "pastebin"
services like
>> http://paste.debian.net/ or http://paste.org.ru/ let us browse the code
>> right away, without having to look at the "Your download will
start
>> shortly..." screen, but don't host *.RData or more than one
file per
>> link. I guess that some other code hosting service like
>> SourceHut/GitHub/GitLab/Gitea/... could fulfil both criteria.
> 
> 
> I would take one step back, and suggest that if any project has as many
> scripts as this one, it should be put in an R package.? You can
> distribute that by building it into a tar.gz file and putting that
> single file in a place where people can download it.? Github and R-forge
> work well for packages, and I assume other hosting services do too,
> though I haven't used many.
> 
> People seem to be unreasonably reluctant to put their code into packages.
> 
> Duncan Murdoch
>