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