Re: C UDF

Hey Imad, You don’t need to set the properties manually when using BUNappend; it will do it for you. Your implementation looks correct, although from a performance perspective I would offer you two tips: - There is no need to allocate/free on every iteration, you can simply create a reasonably sized buffer once and use it for every iteration. A year can never be more than 10~ characters anyway. - In the same vein, using snprintf to determine the length of the string on every iteration is a bit overkill. Mark

On 02 Jan 2017, at 17:31, imad hajj chahine wrote:

Thank you Mark,

Actually I managed to solve this issue on late Friday night, even when using TLoc with integer values i was having random errors in the log and the db was shutdown. Do I need to set tonil and tnil flags when using BATloop/bat_iterator/BUNappend?

Find bellow the complete implementation:

str UDFyearbracket(str *ret, const date *v) { if (*v == date_nil) { *ret = GDKstrdup(str_nil); } else { int y = 0; fromdate(*v, NULL, NULL, &y);

*ret = (str)GDKmalloc(snprintf(NULL, 0, "%d", y) + 1); if (*ret == NULL) throw(MAL, "UDF.yearbracket", "memory allocation failure"); sprintf(*ret, "%d", y); } return MAL_SUCCEED; }

str UDFBATyearbracket(bat *ret, const bat *bid) { BAT *b, *bn; BATiter bi; BUN i,n;

if ((b = BATdescriptor(*bid)) == NULL) throw(MAL, "UDF.BATyearbracket", "Cannot access descriptor"); n = BATcount(b);

bn = COLnew(b->hseqbase, TYPE_str, BATcount(b), TRANSIENT); if (bn == NULL) { BBPunfix(b->batCacheid); throw(MAL, "UDF.BATyearbracket", "memory allocation failure"); }

bi = bat_iterator(b); BATloop(b, i, n) { char *y = NULL; const date *t = (const date *) BUNtail(bi, i); if (*t == date_nil) { y = GDKstrdup(str_nil); } else UDFyearbracket(&y, t); if (BUNappend(bn, y, 0) != GDK_SUCCEED) { goto bailout; } GDKfree(y); }

BBPkeepref(*ret = bn->batCacheid); BBPunfix(b->batCacheid); return MAL_SUCCEED;

bailout: BBPunfix(b->batCacheid); BBPunfix(bn->batCacheid); throw(MAL, "UDF.BATyearbracket", MAL_MALLOC_FAIL); }

Thank You.

On Mon, Jan 2, 2017 at 5:58 PM, Mark Raasveldt mailto:m.raasveldt@cwi.nl> wrote: Hey Imad,

Apologies, scrolling back I noticed that was actually your first attempt at writing the UDF. The source of your error is not encoding related, the error is misleading.

The problem is that in your bulk version you are using Tloc(bn, i) to assign to a string column. Tloc should only be used with constant-sized columns, such as integers or dates. For variable-sized columns such as strings, you should use BUNappend to add values to the column. The reason for that is that string columns are not stored as an array of character pointers, which your initial implementation assumes. Instead, string columns use integers to point into a heap of strings. You are assigning a pointer to one of these integers, which makes MonetDB think the strings are in some random part of your memory. There’s a high chance that that random part of memory does not contain a valid UTF-8 string, hence you get the encoding error.

Try the following bulk implementation instead, using BUNappend instead of Tloc to assign to your column.

Regards,

Mark

str UDFBATyearbracket(bat *ret, const bat *bid) { BAT *b, *bn; BUN i,n; const date *t;

if ((b = BATdescriptor(*bid)) == NULL) throw(MAL, "UDF.BATyearbracket", "Cannot access descriptor"); n = BATcount(b);

bn = COLnew(b->hseqbase, TYPE_str, BATcount(b), TRANSIENT); if (bn == NULL) { BBPunfix(b->batCacheid); throw(MAL, "UDF.BATyearbracket", "memory allocation failure"); } bn->tnonil = 1; bn->tnil = 0;

t = (const date *) Tloc(b, 0); for (i = 0; i < n; i++) { if (*t == date_nil) { BUNappend(bn, str_nil, FALSE); bn->tnonil = 0; bn->tnil = 1; } else { char* ret; UDFyearbracket(&ret, t); BUNappend(bn, ret, FALSE); } t++; }

BATsetcount(bn, n);

bn->tsorted = BATcount(bn)<2; bn->trevsorted = BATcount(bn)<2;

BBPkeepref(*ret = bn->batCacheid); BBPunfix(b->batCacheid); return MAL_SUCCEED; }

----- Original Message ----- From: "Mark Raasveldt" mailto:m.raasveldt@cwi.nl> To: "users-list" mailto:users-list@monetdb.org> Sent: Monday, January 2, 2017 4:32:32 PM Subject: Re: C UDF

Hey Imad,

One of the nice things about UTF-8 is that normal ASCII characters are valid UTF-8. Hence “normal strings” in C are already valid UTF-8. Try simply returning the output from sprintf, like this:

str UDFyearbracket(str *ret, const date *v) { if (*v == date_nil) { *ret = GDKstrdup(str_nil); } else { int year; char *buf; fromdate(*v, NULL, NULL, &year); buf = (char *) GDKmalloc(15); sprintf(buf, "%d", year); *ret = buf; } return MAL_SUCCEED; }

Regards,

Mark

...

On 29 Dec 2016, at 14:35, imad hajj chahine mailto:imad.hajj.chahine@gmail.com> wrote:

Hi Sjoerd,

I tried to used iconv with no luck, I am getting always empty string. I assumed the encoding that i am getting from sprintf are in "ISO-8859-1" Can you please take a look at the following implementation:

str UDFyearbracket(str *ret, const date *v) { if (*v == date_nil) { *ret = GDKstrdup(str_nil); } else { iconv_t cv = iconv_open("UTF-8", "ISO-8859-1"); int factor = 4; size_t fromlen, tolen;

int year; char *buf; char *retChar = (char *)*ret; fromdate(*v, NULL, NULL, &year); buf = (char *) GDKmalloc(15); sprintf(buf, "%d", year);

fromlen = strlen(buf); tolen = factor * fromlen + 1; retChar = (char *) GDKmalloc(tolen); iconv(cv, &buf, &fromlen, &retChar, &tolen); iconv_close(cv); } return MAL_SUCCEED; }

Thanks

_______________________________________________ users-list mailing list users-list@monetdb.org mailto:users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list https://www.monetdb.org/mailman/listinfo/users-list _______________________________________________ users-list mailing list users-list@monetdb.org mailto:users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list https://www.monetdb.org/mailman/listinfo/users-list

_______________________________________________ users-list mailing list users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list