Over in the VO camp such things are "solved" via their (quite amazing) concept of replaceable data drivers (RDD):
in theory it allows to use the .csv or other text files like a .dbf. Cannot give you perf data, as I was too busy with other stuff to follow xSharp dev in the last weeks. But it is an interesting concept totally unknown to us pure fox-heads.

Also in sporadic reading came across a mention that perf of reading csv and similar files is something Julia is excellent in, but have not followed up - but the specific targets of Julia should be up your ally.

regards
thomas

>One issue I still have with VFP speed is reading large text files speedily. Append from *.csv delimited and copy to delim are both indeed pretty impressive even in the face of what data science tools can provide. But once you resort to pure VFP code to read or write text resources on a "line per line" basis, you feel you should possibly write a fll for this and of course you do not!
>
>Of course, I understand that moving from the lib to vfp and back has a processing cost. Do you feel that your faststring.dll could help here in both text file reading and writing?
>
>>>'ve always wanted a CSV class that would let me add rows, access data by the equivalent of a RECNO() and by field names or numbers. Might be interesting to have a FCSV() function which is a cross between FFILETOSTR() and something like a multi-row FALINES(), in that it creates a map into the CSV data that can be accessed easily bow and column / field name. You could add to it, delete from it, operate on it basically like a table.
>
>Looks like a great idea. But I understand it may takes time to deliver!
>
>Thanks for the great thinking. I look forward to your first delivery:-)
>
>Daniel