Re: xlsread in Octave 3.6.4

[Philip Nienhuis]

Markus wrote:
> Am 2013-11-02 21:11, schrieb Philip Nienhuis:
> > Hi Markus,
> >
> > Markus wrote:
> > > <snip>
> > >
> > > I've pushed everything to my github repo[1] again (just the relevant
> > > files for io).
> > > hopefully i don't broke it now (That is quite possible) because i copy
> > > and paste a lot.
> > > Furthermore i've added some nice benchmarks at the end of the
> > > README.md[2]
> > >
> > > cheers
> > > markus
> > >
> > >
> > > [1] https://github.com/markuman/xlsxread
> > > [2] https://github.com/markuman/xlsxread#benchmark
> >
> > It seems you didn't fix the latest sources that I committed but rather
> > patched a local source?
> > But never mind, I again reformatted to Octave coding standards (wasn't
> > much work) and re-added the xls.limits assignment needed to avoid cell
> > errors in xls2oct.
> > I also had to (again) fix undefined references in __OCT_xlsx2oct__.m:
> > L. 117, 126: refs to "@__colnr", should be to "@__col_str_to_number" -
> > do you still have __colnr.m in your source tree?
>
> uh, maybe you're right. i develope with single file xlsxread.m and i
> guess i made often copy & paste errors.
>
> > As regards your remark "setup is a nightmare": All true.
> > But I made a function to make it easier for users, did you catch that?
> > (chk_spreadsheet_support.m); it checks all interfaces and adds all
> > required Java classes to the javaclasspath. All you have to do is
> > download them, put them together into one subdir and tell it to
> > chk_spreadsheet_support.m.
>
> yes, i did it with chk_spreadsheet_support. refering to
> https://wiki.archlinux.org/index.php/Octave#Reading_Microsoft_Excel_Spreadsheets
>
> but xlsread failed as long as JExcelAPI .jar file was added to java
> path. even when i force another interface.

Hey.... I thought I had fixed that bug a while ago... I'll check later.

<snip>
> > Before hitting "Send": I'm really impressed. Your xlsxread stuff is
> > really blazingly fast. testOOXML sheet #2 takes 2.5 secs with OCT,
> > 12.2 with POI, 34+ with UNO (LibreOffice 4.1.2) (all on WinXP)
>
> and matlab run out of memory (on linux 8GB ram). this is the most
> impressiv part for me.

On my Win7 64b 8 GB RAM box, Octave & COM (=Excel 2013) takes just 6,5 
secs to read example.xlsx. I can't test xlsxread (I have no Octave-3.6.4 
installed there yet); but UNO (LibreOffice 4.0) took a leisurely 780 
secs (15 mins). (Java) POI bailed out (a bug in old io-1.2.2 I think).

> next weeks i'll try implement to read just a range from xlsx files.

I figured that a temporary kludge could be to just read all data 
(xlsxread is fast enough), and then just select the requested rectangle 
from the full data matrix.

BYW I think reading strings would also be nice. I already had a brief 
look on how to do that (it is a 2 or 3-step procedure as the strings are 
kept in some other sheet). Vectorizing the procedure will be the real 
challenge.

The io package spreadsheet I/O stuff supports a SPSH_OPTS argument 
(actually a struct with fields that can be true (1) or false (0)). For 
strings we can add a new field "oct_get_strings" or so and in the 
_OCT_xlsx2oct__.m file we can have some if-endif's that test it and if 
"yes"/true also read the strings. That way we can keep the current speed 
if only numeric data are requested.

Tomorrow I'll (try to) make the 1.2.4 release.

Cheers,

Philip