Problem: https://github.com/LimeSurvey/LimeSurvey/blob/master/application/helpers/expressions/em_core_helper.php#L373

GetJavaScriptEquivalentOfExpression() does not check for arguments type mismatch.

Here: https://github.com/LimeSurvey/LimeSurvey/blob/master/application/helpers/expressions/em_core_helper.php#L1323

@DenisChenu On vacation, still?

Coming back now …

The didn't test for "arguments type mismatch" is needed. Because with single choice answer : it's a code, the "1" is a string … forcing to numeric if comparaison is done with a number

PHP compare is done here : https://github.com/LimeSurvey/LimeSurvey/blob/master/application/helpers/expressions/em_core_helper.php#L288

Care to join me on IRC?

Eating before , come on irc this afternoon. I test with the survey

Great! :) A customer had this problem on LimeService, and I've confirmed it. We need to generate type mismatch code for the JS part too, not only PHP.

I will look into how to unit-test JavaScript together with PHPUnit.

It's not empty, it's space … tehn there are a mismatch type with PHP => false , but return true in JS … only with space or with empty string too ?

The space is not a problem.

JS: 1 >= " " ---> true

PHP: false >= " " ---> false

Why "false" and not 1? Because of type mismatch check.

JS: 1 >= "!" or 1 >= "A" or 1 <= "!" or 1 <= "A" ---> false (even if char code is lesser). Maybe a javascript in 2.6lts

See new lss http://demonstration.sondages.pro/563168?newtest=Y

See Using the Equality Operators on https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Comparison_Operators

When comparing a number and a string, the string is converted to a number value. JavaScript attempts to convert the string numeric literal to a Number type value. First, a mathematical value is derived from the string numeric literal. Next, this value is rounded to nearest Number type value.

You're overthinking this. The type mismatch transform

1 >= " "

into

false >= " "

Dump in RDP_EvaluateBinary() for details.

Maybe more relevant: When constructing a condition, the empty string is used to represent "No answer".

RDP_EvaluateBinary : result is an array : 0: the result, 1: don't remind and 2: the value type for EM next function

And we already have a specific fix for JS comparaison for le (i don't add it, surely since 1.92)
https://github.com/LimeSurvey/LimeSurvey/blob/master/application/helpers/expressions/em_core_helper.php#L347

See another updated lss file. tested on 2.67.3
q1.NAOK >= "" : {q1.NAOK >= ""}

q1.NAOK <= "" : {q1.NAOK <= ""}

q1.NAOK == "" : {q1.NAOK == ""}

q1.NAOK >= " " : {q1.NAOK >= " "}

q1.NAOK <= " " : {q1.NAOK <= " "}

q1.NAOK >= "A" : {q1.NAOK >= "A"}

q1.NAOK >= "1A" : {q1.NAOK >= "1A"}

q1.NAOK >= "!" : {q1.NAOK >= "!"}

q1.NAOK <= "A" : {q1.NAOK >= "A"}

q1.NAOK <= "1A" : {q1.NAOK >= "1A"}

q1.NAOK <= "!" : {q1.NAOK >= "!"}

Try with -1 on q1 : same error with " ", but seems there are an issue with "" (emty string) too … �

Some other approaches:

1. Dump _SESSION for each user in JSON in database, BEFORE it's manipulated by the EM. This is to always have a fallback. New field in SurveyDynamic and table survey_xxx called "session_dump"?
2. If relevance = 0 in PHP BUT we have session value, show error (we know there's a mismatch between PHP and JS)
3. Send relevance eq to PHP using Ajax instead of generating JS.

I think there are 2 bugs here :).

1st : no answer IS "" not a space (test with is-empty for example)
2nd : PHP vs JS comparaison : i'm totally unsure what JS do with -1 > "A" ; -1 < "A" … when fix the old issue : information on comment are 'js send always false if one is NUMBER and other is a string without numeric character.

1. OK, then the condition designer if wrong, because it generate NAOK >= " " (space)

1. I think yes :)

PS : the test done is … strange … and there are no valide answer. -1 < no-answer or not ?
0 == no answer or not ?
etc …

Valid answer or not is not important. Only thing that matters is that PHP and JS does the same thing.

Yes :).

1st part : make a complete test with pure javascript to test (for example) 1 < "A" , 1 < " ", 1 < "!" etc … if "number" : js cast it as string (sure at 99.9942%)
2nd part : validate when EM cast as number ( how to : usage of your test ?) think it's ONLY for value from user (and Equation) but unsure for function
3rd part : mimic js in PHP

What do you think of this step ?

Some input from stackoverflow: https://stackoverflow.com/questions/45858228/evaluate-binary-operations-in-both-php-and-javascript

-1 < 'A' ---> false in JavaScript

but

var v = -1
v.toString() < 'A' ---> true in JavaScript

About 1st part: That's basically what I'm doing already in my unit tests.

You think the problem is that JavaScript casts numbers to strings at certain points, and PHP does not?

Chapter about type coersion in JavaScript: https://www.safaribooksonline.com/library/view/you-dont-know/9781491905159/ch04.html

Another problem: EM in PHP does not know at evaluation if the function is onlynumber. Look at this:

"0" == " "

EM will treat this as is in PHP, "0" == " ", which is false, but in JS LEMval("0") will return 0, not "0", because onlynum = 1, and 0 == " " is true in JS.

-1 < 'A or 1 < 'A' or 1 <= 'A' or … -1 <= '!' => always false. It's the reason of " "arguments type mismatch" part … because PHP cast as string by default …

excet with space … (and empty : empty is 0). Seems javascript set " " to "" then to 0 …

Sure. I was talking about the logic in LEMval(), which will cast "0" to 0, but this doesn't happen in PHP. Basically, I think EM needs to know about the onlynum property, too.

Another idea: Scrap <, >, <= etc and replace them with LEMcompare(operation, type, arg1, arg2) to get precise control over type-casting.

0 == " " : yes seems when using " " , javascript just trim it

Maybe we must treat like '' and '0' , see https://github.com/LimeSurvey/LimeSurvey/blob/master/application/helpers/expressions/em_core_helper.php#L370

Then this fix THIS issue ;)

Yes, I will try that, might fix it for ==.

Something like :

$bNumericArg2 = !trim($arg2[0]) || strval(floatval($arg2[0]))==strval($arg2[0]); https://github.com/LimeSurvey/LimeSurvey/blob/c21c16f990894859626efe28f0749f727b7a4a90/application/helpers/expressions/em_core_helper.php#L287

But need test with " 0 " in js.

Someone already use === in expression ?

My tests run JS too, just tell me which expression to add.

Maybe :
array('0',1,'WORD')
array('0',1,'NUMBER')
array('0',1,'STRING')

'0' : replace by [space] , '!' (before number in ascii), 'A' (after number in ascii), '0xA' (numeric in PHP not in js), '001.100' (DB and 'maybe' lemval under condition).

comparaison
'!' >= 1
'A' <= 1
' ' <= 1
' ' <= 1
'-1' <= 1
'001.100' + 1 (can give 2 or '001.1001', depend of STRING or not
'001.100' - 1 (0 or NAN ?)

A lot can be added …

'0xA' with 17 and 15 : but i think there are an issue currently

Does this issue only appear for numerical questions?

The answer is no, I can reproduce it with short text question, too.

I think this is not possible to solve without knowing the question type in EM core, if it's numerical or not, since LEMval() behaves different in that case. So how to add that context in EM core?

EM core : WORD or NUMBER if i don't make error …

NUMBER == isnumeric in js.

BUT : LEMval do : "1" to 1 with any question type not only for numeric question type.

I get all WORD, even for numerical type.

@ollehar : i work on string/numeric comparaison currently https://bugs.limesurvey.org/view.php?id=14337#c52311

Seems PHP need something like this:

            case '<':
            case 'lt':
                if ($bMismatchType) {
                    $result = array(false, $token[1], 'NUMBER');
                 } elseif ($bBothString && !$bBothNumeric) {
                    $result = array(strcmp($arg1[0],$arg2[0]) < 0 , $token[1], 'NUMBER');
                } else {
                    $result = array(($arg1[0] < $arg2[0]), $token[1], 'NUMBER');
                }
                break;

Maybe it can fix this part too.

Phew... I don't really have the head space to dig into this, but I can help you run the EM tests I created for this bug. It's marked as skipped right now. You need to install NodeJS to run JavaScript without the browser.

PS : i do a pull request but only next week.

1. Adding strcmp($arg1[0],$arg2[0]) when the 2 are string
2. Adding a «forced string » check in https://github.com/LimeSurvey/LimeSurvey/blob/7db9677aa40de2c6acc149e41917bd2d29aa705c/application/helpers/expressions/em_core_helper.php#L285 (issue with "3A" < 5+"")

I'd recommend to NOT change anything before having a proper test suite. The chances of breaking something else is very big. :|

Yes, i know it's the reason why i only fix '5' < '20' in https://bugs.limesurvey.org/view.php?id=14337

A precision about type, currently there are, for PHP

1. WORD : user entered value or answer code : Q00.NAOK. Can be set as number if it can be number. This behaviour was in https://bugs.limesurvey.org/view.php?id=8324. See my point of view here : https://bugs.limesurvey.org/view.php?id=8324#c27049 (and the related issue)
2. NUMBER : result of a number function (sum) OR Q00.NAOK with a numeric type question (to check)
3. STRING : result of a string function (join ?)
4. DQ_STRING and DS_STRING : "a" and 'a'. To mimic JS 2+"" , there are the forced string test