NeoCSVReader and an empty field at the very end of a file
Hi there, I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special. It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this: SecondLastColumnValue;;<EOF> In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF. So I changed readField to this: readField ^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]] and all seems fine so far. Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation. Joachim -- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Hi, I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator: readSeparator ^self atEnd ifFalse: [self peekFor: separator] As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix? Joachim Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Hi Joachim, First, thanks for the feedback. Second, since you are on a different platform, that might be a factor. Did you test your problem on Pharo itself ? Because there are already unit tests specifically for the case you describe: #testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted These obviously pass for Pharo, do they pass for you ? Maybe your problem case is slightly different though ? Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Hi Sven, I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago. So it is very likely I was bitten by platform differences. Thanks for answering. Joachim Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd. It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
I am not complaining. It just makes porting harder. What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side... Joachim Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Joachim, which results do you get on VA Smalltalk for the tests? Do they all pass? --Hannes On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth. Joachim Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Le 01/07/2015 21:07, jtuchel@objektfabrik.de a écrit :
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Maybe we should write a refactoring to change those curly braces into an equivalent Array based syntax? I'll have some significant porting from Pharo to do one of these days, so maybe I'll setup something for that. Thierry
Joachim
Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- -----------------------------------------------------------------------
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
You mean RewriteRules? I was planning to do that, for this exact reason. But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time... Joachim Am 01.07.15 um 21:10 schrieb Thierry Goubier:
Le 01/07/2015 21:07, jtuchel@objektfabrik.de a écrit :
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Maybe we should write a refactoring to change those curly braces into an equivalent Array based syntax?
I'll have some significant porting from Pharo to do one of these days, so maybe I'll setup something for that.
Thierry
Joachim
Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- -----------------------------------------------------------------------
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- -----------------------------------------------------------------------
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Le 01/07/2015 21:16, jtuchel@objektfabrik.de a écrit :
You mean RewriteRules? I was planning to do that, for this exact reason. But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that... Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is. Thierry
Hi thierry mark will arriv at Lille saturday for a month. He is working with Camille on a new tree pattern matching algorithm because we lost energy on RB matcher. Stef Le 1/7/15 21:29, Thierry Goubier a écrit :
Le 01/07/2015 21:16, jtuchel@objektfabrik.de a écrit :
You mean RewriteRules? I was planning to do that, for this exact reason. But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that...
Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is.
Thierry
Hi Stef, Le 02/07/2015 23:18, stepharo a écrit :
Hi thierry
mark will arriv at Lille saturday for a month. He is working with Camille on a new tree pattern matching algorithm
A generic algorithm or one tied to the RB AST?
because we lost energy on RB matcher.
What was your analysis? Thanks, Thierry
Stef
Le 1/7/15 21:29, Thierry Goubier a écrit :
Le 01/07/2015 21:16, jtuchel@objektfabrik.de a écrit :
You mean RewriteRules? I was planning to do that, for this exact reason. But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that...
Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is.
Thierry
Hi thierry
mark will arriv at Lille saturday for a month. He is working with Camille on a new tree pattern matching algorithm
A generic algorithm or one tied to the RB AST? A generic algo. We will talk with Camille monday
because we lost energy on RB matcher.
What was your analysis?
JC did his Phd on code transformation and with anne they try to really document how the unification works in RB and ... after tears and pain they gave up. So we really want to see if we cannot do something better.
Thanks,
Thierry
Stef
Le 1/7/15 21:29, Thierry Goubier a écrit :
Le 01/07/2015 21:16, jtuchel@objektfabrik.de a écrit :
You mean RewriteRules? I was planning to do that, for this exact reason. But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that...
Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is.
Thierry
Le 4 juil. 2015 10:58 PM, "stepharo" <stepharo@free.fr> a écrit :
Hi thierry
mark will arriv at Lille saturday for a month. He is working with Camille on a new tree pattern matching algorithm
A generic algorithm or one tied to the RB AST?
A generic algo. We will talk with Camille monday
One more on my list of things to follow.
because we lost energy on RB matcher.
What was your analysis?
JC did his Phd on code transformation and with anne they try to really document how the unification works in RB and ... after tears and pain they gave up.
Did they asked John Brant? He told me he was using the RB parser for everything Smalltalk and SmaCC for other languages. Oh, for JC: SmaCC does both string and ast node replacement... With a small bug in SmaCCString I need to correct (but with a design suitable for doing that over very long files).
So we really want to see if we cannot do something better.
Or something you master ;) Thierry
Thanks,
Thierry
Stef
Le 1/7/15 21:29, Thierry Goubier a écrit :
Le 01/07/2015 21:16, jtuchel@objektfabrik.de a écrit :
You mean RewriteRules? I was planning to do that, for this exact
reason.
But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that...
Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is.
Thierry
Check the chapter Anne and JC wrote on pharoinprogress. We discussed a lot about RB because the unification does not identify the subtree people think about. I will let Camille and JC comment on it. What we saw is that we should try something because they are too many edge corner. Le 5/7/15 01:12, Thierry Goubier a écrit :
Le 4 juil. 2015 10:58 PM, "stepharo" <stepharo@free.fr <mailto:stepharo@free.fr>> a écrit :
Hi thierry
mark will arriv at Lille saturday for a month. He is working with Camille on a new tree pattern matching algorithm
A generic algorithm or one tied to the RB AST?
A generic algo. We will talk with Camille monday
One more on my list of things to follow.
because we lost energy on RB matcher.
What was your analysis?
JC did his Phd on code transformation and with anne they try to really document how the unification works in RB and ... after tears and pain they gave up.
Did they asked John Brant? He told me he was using the RB parser for everything Smalltalk and SmaCC for other languages.
Oh, for JC: SmaCC does both string and ast node replacement... With a small bug in SmaCCString I need to correct (but with a design suitable for doing that over very long files).
So we really want to see if we cannot do something better.
Or something you master ;)
Thierry
Thanks,
Thierry
Stef
Le 1/7/15 21:29, Thierry Goubier a écrit :
Le 01/07/2015 21:16, jtuchel@objektfabrik.de
<mailto:jtuchel@objektfabrik.de> a écrit :
You mean RewriteRules? I was planning to do that, for this exact
reason.
But you know how things go when there is a long list of priorities. I know it saves a lot of time and makes porting so much more safe, but it needs an up-front investment in time...
Yes, rewrite rules. Mark Rizun GUI attempts are a very interesting direction to handle that...
Oh, by the way, Mark, if you're listening... I did what you were trying to do by changing AST nodes positions in Pharo, but on a SmaCC based AST for another language. I have to admit that you were right to try to do it that way, but that the Pharo/RB AST infrastructure wasn't suitable whereas SmaCC infrastructure is.
Thierry
Le 05/07/2015 12:41, stepharo a écrit :
Check the chapter Anne and JC wrote on pharoinprogress. We discussed a lot about RB because the unification does not identify the subtree people think about. I will let Camille and JC comment on it. What we saw is that we should try something because they are too many edge corner.
I'm certainly interested to learn what is happening... In part because I need matching on code with more knowledge than the AST. Thanks in advance, Thierry
Le 1/7/15 21:10, Thierry Goubier a écrit :
Le 01/07/2015 21:07, jtuchel@objektfabrik.de a écrit :
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Maybe we should write a refactoring to change those curly braces into an equivalent Array based syntax?
I'll have some significant porting from Pharo to do one of these days, so maybe I'll setup something for that. Why from Pharo? What is the problem?
Stef
Le 02/07/2015 23:16, stepharo a écrit :
Le 1/7/15 21:10, Thierry Goubier a écrit :
Le 01/07/2015 21:07, jtuchel@objektfabrik.de a écrit :
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Maybe we should write a refactoring to change those curly braces into an equivalent Array based syntax?
I'll have some significant porting from Pharo to do one of these days, so maybe I'll setup something for that. Why from Pharo? What is the problem?
Not really any problem. There are just a thing or two I'd like to try, and I need to get up to speed on refactorings. Thierry
But before you ask for help on the Pharo ML you should run the NeoCSV code on Pharo to validate that your assumption hold there. If not, it is a porting error/problem. And unit tests are *VERY IMPORTANT* to maintain our sanity across versions and/or platforms. Again, what you report as a potential problem isn't one, I think. Please test on Pharo itself.
On 01 Jul 2015, at 21:07, jtuchel@objektfabrik.de wrote:
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Joachim
Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Sven, you are, of course, right! I should have checked in Pharo first. But since my version is old anyways, I would have to try with at least two versions in order to make sure if the bug is fixed or not. As long as keeping ports current is a lot of work due to (increasing) differences between Phar and most other dialects, this is hard work ;-) I should invest in it anyways, however, since I am using your code from Pharo and you didn't ask me to use it... So, to address this Array construction issue (as an example of others). This cannot be handled with Grease or the like, since what is in my way is the Smalltalk compiler (I am trying to concince Instantiations to support {} for literal Array construction, but until I'm successful...). Given somebody wrote RewriteRules to replace {} with "(Array new: x) addAll:", this should probably be used on non-Pharo-platforms as part of a port, so that the "original" code is still the Metacello version it was on Pharo. My plan for a port: 1. download .mcz in its "original" version and don't modify it 2. import and modify on target platform 3. test and publish on target platform's public repository What do you think would be a good process to porting when it involves such steps? The problem with Step 2 is: I am not sure if RewriteRules can work in code that's not compiled. (Well, I am mostly sure this won't work, just asking for confirmation) You cannot compile methods with Arrays constructed with curly braces on VAST. So another possible process would be: 1. use Pharo to modify downloaded code in order to make it importable on target platform 2. publish this code somewhere (not official Pharo repositories, maybe StHub) 3. Download the modified version and import into target dialect 4. Make final adjustments to make code and tests run 5. Publish to target platform's public repository Any thoughts or ideas? People interested in joining efforts? And, now that we're on a new topic anyways, misusing this thread (mea culpa): I still wonder how we could make it easier to contribute back. Every time I find a bug in some code from Pharo, there are several things to consider: 1. Is there a newer version on Pharo that doesn't have this bug? a) yes: port newer version b) no: I'm in trouble - see below in case of 1b), there are two options now: 1. I fire up Pharo and fix the bug in the original code, then commit the changes to the official repository/ies, and then port the fixed version 2. I contact the ML or the author/maintainer of the code and give them a bug report and hopefully a fix, which may or may not be compatible with Pharo. Then I can either port the fixed version or just keep my fixed version (if the fix is the same, this is not a problem from the code perspective, but maybe from a version numbering standpoint). I am interested in ideas here and would like to hear from people interested in discussing these issues. Maybe ESUG would be a good opportunity to start discussing. Maybe we can start a new initiative for this problem area... Joachim Am 01.07.15 um 21:27 schrieb Sven Van Caekenberghe:
But before you ask for help on the Pharo ML you should run the NeoCSV code on Pharo to validate that your assumption hold there. If not, it is a porting error/problem.
And unit tests are *VERY IMPORTANT* to maintain our sanity across versions and/or platforms.
Again, what you report as a potential problem isn't one, I think. Please test on Pharo itself.
On 01 Jul 2015, at 21:07, jtuchel@objektfabrik.de wrote:
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Joachim
Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
On Thu, Jul 2, 2015 at 3:48 PM, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
Sven,
you are, of course, right! I should have checked in Pharo first. But since my version is old anyways, I would have to try with at least two versions in order to make sure if the bug is fixed or not. As long as keeping ports current is a lot of work due to (increasing) differences between Pharo and most other dialects, this is hard work ;-) I should invest in it anyways, however, since I am using your code from Pharo and you didn't ask me to use it...
This reminds me of Joel Spolsky's article on "barriers to entry." He makes an interesting point with his "PayMyBills" example that a hidden barrier to entry is not how hard it is to switch in, but how hard it is to switch out [or port]. cheers -ben
Hello Joachim In May I did a port of NeoCSV to Cuis Smalltalk. I followed a process like this 1. download .mcz in its "original" version and unpack it to have the st file only. Keep the Core and the Tests st file in the target repository for reference. 2. Make a copy of the two files 3. Modify the st files with search/replace operations for obvious and safe changes 4. Import and further modify on target platform 5. Run tests and resolve remaining issues 6. Publish on target platform's public repository 7. Document the process so that it can be redone from scratch much faster by myself or somebody else in a semi-automatic way. Remarks: 1) https://github.com/hhzl/Cuis-NeoCSV/blob/master/Neo-CSV-Core-orig.st is 1011 LOC https://github.com/hhzl/Cuis-NeoCSV/blob/master/Neo-CSV-Tests-orig.st is 805 LOC It took me two attempts to arrive at the goal https://github.com/hhzl/Cuis-NeoCSV In the first attempt nearly all tests failed. But this was because of some minor differences which caused most of the problems. After having them fixed the number of tests which passed raised quickly. In the end it needed a combination of systematic changes and some 'compatibility layer' on Cuis which is necessary for the port. The package used for that contains much more than needed, I did not take the time yet to figure out which methods actually are needed. I think only very few methods are needed. I think the Neo-CSV package is very useful to do workspace based applications for an explorative analysis of complex data. CSV data is a representation of one of the most common file formats (Excel data). And we need to be able to do this in various Smalltalk distributions. The Neo-CSV package is well written and not all that complex. Porting can be done in some hours. Sven was very helpful and added another test for me to illustrate a particular use, actually my main use case. Summary: a) So the porting experience was good. b) Porting the tests is a necessity to prove that the port has been done. c) I think you can do the same thing quickly for VA Smalltalk if you actually take the plunge. Way forward: It might be worthwile to consider to maintain a github repository with steps 1,2 and 3 from above. FileOut format for ease of use. Using another repo will distinguish the effort to be a modified _copy_ of http://smalltalkhub.com/#!/~SvenVanCaekenberghe/Neo/ Actually Sven already has taken precautions to make the code porting-friendly and he might want to further facilitate this. I am considering a port to Amber Smalltalk, but interestingly though Amber is said to be Pharo compatible there are some problems which I have to resolve first. HTH Hannes On 7/2/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
Sven,
you are, of course, right! I should have checked in Pharo first. But since my version is old anyways, I would have to try with at least two versions in order to make sure if the bug is fixed or not. As long as keeping ports current is a lot of work due to (increasing) differences between Phar and most other dialects, this is hard work ;-) I should invest in it anyways, however, since I am using your code from Pharo and you didn't ask me to use it...
So, to address this Array construction issue (as an example of others). This cannot be handled with Grease or the like, since what is in my way is the Smalltalk compiler (I am trying to concince Instantiations to support {} for literal Array construction, but until I'm successful...).
Given somebody wrote RewriteRules to replace {} with "(Array new: x) addAll:", this should probably be used on non-Pharo-platforms as part of a port, so that the "original" code is still the Metacello version it was on Pharo.
My plan for a port: 1. download .mcz in its "original" version and don't modify it 2. import and modify on target platform 3. test and publish on target platform's public repository
What do you think would be a good process to porting when it involves such steps?
The problem with Step 2 is: I am not sure if RewriteRules can work in code that's not compiled. (Well, I am mostly sure this won't work, just asking for confirmation) You cannot compile methods with Arrays constructed with curly braces on VAST.
So another possible process would be: 1. use Pharo to modify downloaded code in order to make it importable on target platform 2. publish this code somewhere (not official Pharo repositories, maybe StHub) 3. Download the modified version and import into target dialect 4. Make final adjustments to make code and tests run 5. Publish to target platform's public repository
Any thoughts or ideas? People interested in joining efforts?
And, now that we're on a new topic anyways, misusing this thread (mea culpa):
I still wonder how we could make it easier to contribute back. Every time I find a bug in some code from Pharo, there are several things to consider: 1. Is there a newer version on Pharo that doesn't have this bug? a) yes: port newer version b) no: I'm in trouble - see below
in case of 1b), there are two options now: 1. I fire up Pharo and fix the bug in the original code, then commit the changes to the official repository/ies, and then port the fixed version 2. I contact the ML or the author/maintainer of the code and give them a bug report and hopefully a fix, which may or may not be compatible with Pharo. Then I can either port the fixed version or just keep my fixed version (if the fix is the same, this is not a problem from the code perspective, but maybe from a version numbering standpoint).
I am interested in ideas here and would like to hear from people interested in discussing these issues. Maybe ESUG would be a good opportunity to start discussing. Maybe we can start a new initiative for this problem area...
Joachim
Am 01.07.15 um 21:27 schrieb Sven Van Caekenberghe:
But before you ask for help on the Pharo ML you should run the NeoCSV code on Pharo to validate that your assumption hold there. If not, it is a porting error/problem.
And unit tests are *VERY IMPORTANT* to maintain our sanity across versions and/or platforms.
Again, what you report as a potential problem isn't one, I think. Please test on Pharo itself.
On 01 Jul 2015, at 21:07, jtuchel@objektfabrik.de wrote:
I gave up on porting tests. VAST doesn't support creating Arrays with curly braces, and they are used all over the place in Pharo code. I wuld use them if I had them, so this is not an accuse or anything. It's just a huge load of work to keep tests up to date when porting newer versions, so I just ignore tests when porting. A sad truth.
Joachim
Am 01.07.15 um 16:14 schrieb H. Hirzel:
Joachim,
which results do you get on VA Smalltalk for the tests?
Do they all pass?
--Hannes
On 7/1/15, jtuchel@objektfabrik.de <jtuchel@objektfabrik.de> wrote:
I am not complaining. It just makes porting harder.
What made the thing especially strange was that I only had that problem with uploaded files because the Browser removes the trailing CrLf (or better, doesn't add another one in the multipart form data). So reading from a file all works like a charm, but not when you process the same, uploaded file on the server side...
Joachim
Am 01.07.15 um 15:12 schrieb Sven Van Caekenberghe:
Yes, on Pharo, #next (and #peek or #peekFor:) all return nil when #atEnd.
It is the way it is (I am personally for stricter semantics), but that fact is certainly used in code all allround the place.
On 01 Jul 2015, at 15:06, jtuchel@objektfabrik.de wrote:
Hi Sven,
I didn't test on Pharo. But I remember seeing differences in the way Pharo and VAST react to reads beyond the end of a Stream. Not sure, but this could have been in NeoCSV context two or three years ago.
So it is very likely I was bitten by platform differences.
Thanks for answering.
Joachim
Am 01.07.15 um 14:05 schrieb Sven Van Caekenberghe:
Hi Joachim,
First, thanks for the feedback.
Second, since you are on a different platform, that might be a factor.
Did you test your problem on Pharo itself ?
Because there are already unit tests specifically for the case you describe:
#testEmptyLastFieldUnquoted #testEmptyLastFieldQuoted
These obviously pass for Pharo, do they pass for you ?
Maybe your problem case is slightly different though ?
Sven
On 01 Jul 2015, at 13:40, jtuchel@objektfabrik.de wrote:
Hi,
I've tried porting SvenVanCaekenberghe.20 and see the same problems in this version. IN addition to the fix already mentioned, I also had to change readSeparator:
readSeparator
^self atEnd ifFalse: [self peekFor: separator]
As far as I can tell by now, this fixes the problem at hand. Any ideas if this is a safe fix?
Joachim
Am 01.07.15 um 12:35 schrieb jtuchel@objektfabrik.de:
Hi there,
I am on VA Smalltalk and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
--
Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
-- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1 http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1
Hannes Hirzel wrote
though Amber is said to be Pharo compatible there are some problems which I have to resolve first.
Pharo is the reference implementation, but is a subset. I guess technically that's "compatible", but "compatible" is vague enough to be easily misleading... ----- Cheers, Sean -- View this message in context: http://forum.world.st/NeoCSVReader-and-an-empty-field-at-the-very-end-of-a-f... Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com.
Joachim, Le 2 juil. 2015 9:49 AM, "jtuchel@objektfabrik.de" <jtuchel@objektfabrik.de> a écrit :
The problem with Step 2 is: I am not sure if RewriteRules can work in
code that's not compiled. (Well, I am mostly sure this won't work, just asking for confirmation) You cannot compile methods with Arrays constructed with curly braces on VAST. Rewrite rules do work on non compiled code. RB contains all the necessary models to do everything RB on non compiled code. Regards, Thierry
Le 1/7/15 12:35, jtuchel@objektfabrik.de a écrit :
Hi there,
I am on VA Smalltalk For your next successful project you should jump in Ph.... :)
and therefor using an older version of NeoCSV (SvenVanCaekenberghe.14). I found a bug in this old version that is somewhat special.
It seems NeoCSV cannot handle the situation where the very last field is just empty AND if there is no trailing CRLF at the end of the file. Somethinng like this:
SecondLastColumnValue;;<EOF>
In that case, readField fails because it tries to do a readQuotedField or readUnquotedField, both of which try to read beyond EOF.
So I changed readField to this:
readField
^self atEnd "In case the very last field of a file is empty, like '45;56;;'" ifTrue: [''] ifFalse: [self peekQuote ifTrue: [self readQuotedField] ifFalse: [self readUnquotedField]]
and all seems fine so far.
Side note: My original file has a trailing CrLf but if I upload it via a browser to a Seaside Server, the Browser cuts off the trailing CrLf (I can see this in the Browser's Network debugging tools - both in IE and FF) - so it seems NeoCSV has to be ready for this situation.
Joachim
participants (7)
-
Ben Coman -
H. Hirzel -
jtuchel＠objektfabrik.de -
Sean P. DeNigris -
stepharo -
Sven Van Caekenberghe -
Thierry Goubier