Loading ...
Sorry, an error occurred while loading the content.

RE: [jslint] Circular Function Definitions

Expand Messages
  • Rob Richardson
    Felix, The primary concern is you re trying to use a function earlier in your code file than when it is defined. If you put var f = function () {... before
    Message 1 of 17 , Dec 31, 2010
    • 0 Attachment
      Felix,

      The primary concern is you're trying to use a function earlier in your code
      file than when it is defined. If you put var f = function () {... before
      function callback(... then JSLint would likely be fine. I understand that
      given the business rules around the code in question that callback() (and
      thus f()) would never be used before it was defined, but JSLint doesn't know
      that.

      The secondary concern is that you're passing in a flag that says whether you
      should use a follow-up function. Perhaps you refactored f() out of
      callback() for readability, perhaps f() is used in more than one place,
      perhaps you never considered putting the contents of f inside callback.
      Alas, the "moreToDo is true means run f()" is a bit confusing. Passing the
      function to run into callback() instead of a flag makes callback() more
      generic, and makes this point a bit more clear, and that was the focus of
      Erik's thoughts.

      Rob


      -----Original Message-----
      From: jslint_com@yahoogroups.com [mailto:jslint_com@yahoogroups.com] On
      Behalf Of Felix E. Klee
      Sent: Friday, December 31, 2010 1:12 PM
      To: jslint_com@yahoogroups.com
      Subject: Re: [jslint] Circular Function Definitions

      On Fri, Dec 31, 2010 at 7:21 PM, Erik Eckhardt <erik@...> wrote:
      > function callback(morefn) {
      > // do something ...
      > if (typeof(morefn) === 'function') {
      > moref();
      > }
      > }
      >
      > f = function () {
      > sendToServer('xyz', callback ? f : null);
      > };

      Are you sure this makes sense? The second last line will always evaluate to:

      sendToServer('xyz', f);

      And I don't see how this solves the issue I raised:

      var f;

      function callback(moreToDo) {
      // ...
      }

      f = function () {
      // ...
      };

      That's inconsistent, and thus confusing. Of course one could write:

      var callback, f;

      callback = function (moreToDo) {
      // ...
      };

      f = function () {
      // ...
      };

      Now things are consistent. But is that the way to go?
    • Felix E. Klee
      On Fri, Dec 31, 2010 at 9:49 PM, Rob Richardson ... No, the opposite is true. See the second example in my original post. That works
      Message 2 of 17 , Jan 1, 2011
      • 0 Attachment
        On Fri, Dec 31, 2010 at 9:49 PM, Rob Richardson <erobrich@...>
        wrote:
        > If you put var f = function () {... before function callback(... then
        > JSLint would likely be fine.

        No, the opposite is true. See the second example in my original post.
        That works fine with JSLint. Changing the order would make JSLint
        complain about "callback" not being defined.

        But that's not really the issue.

        > The secondary concern is that you're passing in a flag that says
        > whether you should use a follow-up function.

        The code is more complicated in reality. There is a lot happening in
        "f", and whether to send another request to the server is determined
        based not just on the state of "moreToDo".

        The code chains XHRs, to download server generated data.

        > Passing the function to run into callback() instead of a flag makes
        > callback() more generic, and makes this point a bit more clear, and
        > that was the focus of Erik's thoughts.

        Erik's code would just generate an infinite loop, because it evaluates
        to:

        f = function () {
        sendToServer('xyz', f);
        };

        But that's *all* besides the point. My question, I think, can be
        summarized to:

        For the sake of consistency, and thus readability, should I always
        define functions as:

        var f;

        // ...

        f = function () {
        // ...
        }

        instead of the short form:

        function f() {
        // ..
        }

        ?

        Furthermore, can this be enforced with JSLint?
      • Felix E. Klee
        On Sat, Jan 1, 2011 at 11:59 AM, Felix E. Klee ... Sorry, that should be: There is a lot happening in callback
        Message 3 of 17 , Jan 1, 2011
        • 0 Attachment
          On Sat, Jan 1, 2011 at 11:59 AM, Felix E. Klee <felix.klee@...>
          wrote:
          > There is a lot happening in "f"

          Sorry, that should be:

          There is a lot happening in "callback"
        • Rob Richardson
          Ah, I see I may have misunderstood. You re proposing the following: var a,b; a = function () { // ... b(); // ... } b = function () { // ... a(); // ... } Is
          Message 4 of 17 , Jan 1, 2011
          • 0 Attachment
            Ah, I see I may have misunderstood. You're proposing the following:

            var a,b;

            a = function () {
            // ...
            b();
            // ...
            }

            b = function () {
            // ...
            a();
            // ...
            }

            Is this the dilemma at hand?

            Rob


            -----Original Message-----
            From: jslint_com@yahoogroups.com [mailto:jslint_com@yahoogroups.com] On
            Behalf Of Felix E. Klee
            Sent: Saturday, January 01, 2011 3:59 AM
            To: jslint_com@yahoogroups.com
            Subject: Re: [jslint] Circular Function Definitions

            On Fri, Dec 31, 2010 at 9:49 PM, Rob Richardson <erobrich@...>
            wrote:
            > If you put var f = function () {... before function callback(... then
            > JSLint would likely be fine.

            No, the opposite is true. See the second example in my original post.
            That works fine with JSLint. Changing the order would make JSLint
            complain about "callback" not being defined.

            But that's not really the issue.

            > The secondary concern is that you're passing in a flag that says
            > whether you should use a follow-up function.

            The code is more complicated in reality. There is a lot happening in
            "f", and whether to send another request to the server is determined
            based not just on the state of "moreToDo".

            The code chains XHRs, to download server generated data.

            > Passing the function to run into callback() instead of a flag makes
            > callback() more generic, and makes this point a bit more clear, and
            > that was the focus of Erik's thoughts.

            Erik's code would just generate an infinite loop, because it evaluates
            to:

            f = function () {
            sendToServer('xyz', f);
            };

            But that's *all* besides the point. My question, I think, can be
            summarized to:

            For the sake of consistency, and thus readability, should I always
            define functions as:

            var f;

            // ...

            f = function () {
            // ...
            }

            instead of the short form:

            function f() {
            // ..
            }

            ?

            Furthermore, can this be enforced with JSLint?
          • Felix E. Klee
            ... Yes. To me it seems reasonable to - for consistency - avoid the short form: function f() { // ... } And I would like to enforce that in the entire project,
            Message 5 of 17 , Jan 3, 2011
            • 0 Attachment
              On Sun, Jan 2, 2011 at 4:54 AM, Rob Richardson <erobrich@...> wrote:
              > var a,b;
              >
              > a = function () {
              > // ...
              > b();
              > // ...
              > }
              >
              > b = function () {
              > // ...
              > a();
              > // ...
              > }
              >
              > Is this the dilemma at hand?

              Yes.

              To me it seems reasonable to - for consistency - avoid the short form:

              function f() {
              // ...
              }

              And I would like to enforce that in the entire project, with JSLint.
              However, as JSLint does not offer a corresponding option, I wonder if I
              am on the wrong track, overdoing things.
            • Erik Eckhardt
              My apologies. My code was written too hastily. I thought you were using a Boolean to indicate whether to make the return call to f. The point I was trying to
              Message 6 of 17 , Jan 3, 2011
              • 0 Attachment
                My apologies. My code was written too hastily. I thought you were using a
                Boolean to indicate whether to make the return call to f.

                The point I was trying to get across is that you don't need the circularity.

                There are many ways to solve this. Here's one: Any time you want something
                to happen after the callback, pass the call to the callback in an anonymous
                function:

                f = function() {
                ___// do stuff
                ___xhr(params, callback); // plain call to cb func
                ___xhr(params, function() {callback(); f();}); // circular call back to f
                }

                Or pass the after-callback function as a parameter to the callback function
                (again through use of an anonymous function).

                or: `callback.returnfn = f; // now check property in that function and call
                if needed

                Or use a library like jquery and define a custom event, then bind all the
                handlers to the event you want, and trigger it any time.

                Circularity isn't required.

                Erik

                On Fri, Dec 31, 2010 at 12:12 PM, Felix E. Klee <felix.klee@...> wrote:

                >
                >
                > On Fri, Dec 31, 2010 at 7:21 PM, Erik Eckhardt <erik@...<erik%40eckhardts.com>>
                > wrote:
                > > function callback(morefn) {
                > > // do something ...
                > > if (typeof(morefn) === 'function') {
                > > moref();
                > > }
                > > }
                > >
                > > f = function () {
                > > sendToServer('xyz', callback ? f : null);
                > > };
                >
                > Are you sure this makes sense? The second last line will always evaluate
                > to:
                >
                > sendToServer('xyz', f);
                >
                > And I don't see how this solves the issue I raised:
                >
                > var f;
                >
                > function callback(moreToDo) {
                > // ...
                > }
                >
                > f = function () {
                > // ...
                > };
                >
                > That's inconsistent, and thus confusing. Of course one could write:
                >
                > var callback, f;
                >
                > callback = function (moreToDo) {
                > // ...
                > };
                >
                > f = function () {
                > // ...
                > };
                >
                > Now things are consistent. But is that the way to go?
                >
                >


                [Non-text portions of this message have been removed]
              • Felix E. Klee
                ... No, that was correct. In the example code, I *was* using a boolean. Only in the real code, things are more complex: Based on what the XHR returns some
                Message 7 of 17 , Jan 4, 2011
                • 0 Attachment
                  On Tue, Jan 4, 2011 at 7:53 AM, Erik Eckhardt <erik@...> wrote:
                  > I thought you were using a Boolean to indicate whether to make the
                  > return call to f.

                  No, that was correct. In the example code, I *was* using a boolean. Only
                  in the real code, things are more complex:

                  Based on what the XHR returns some calculations are performed, and - if
                  needed - another XHR is made.

                  > The point I was trying to get across is that you don't need the
                  > circularity.

                  What's bad about the circularity?

                  > xhr(params, function() {callback(); f();}); // circular call back to f

                  That would again be an infinite loop. The body of the anonymous function
                  would need to be at least slightly more complex. I don't think this is
                  increases readability.
                • Erik Eckhardt
                  If your example isn t the full story, please post an example of the full story! One thing that s wrong with the circularity is the cross-dependence as JSLint
                  Message 8 of 17 , Jan 4, 2011
                  • 0 Attachment
                    If your example isn't the full story, please post an example of the full
                    story!

                    One thing that's wrong with the circularity is the cross-dependence as
                    JSLint is showing you. It's not a huge deal, but an annoying one, perhaps.

                    In any case I thought you might find it useful to use some of the techniques
                    I mentioned, where if a function determines that an additional step is
                    needed, it manages it within the function rather than making the callback
                    function have a hard coded reference back. The scheme you're using now seems
                    to make sense but as your application grows you may find that it becomes
                    unworkable. For example, what if you eventually have two functions that may
                    or may not need to be called after an xhr? Are you going to pass two
                    booleans? Or God forbid, a code indicating which functions to run? If from
                    the start you simply passed around single callback functions (which
                    themselves could be a string of functions) this problem would be solved.

                    function myxhr(url, params, callback) {
                    ___//do stuff;
                    doXhr(function() {myxhrreturn(xmlhttp, callback):});
                    }

                    function myxhrreturn(xmlhttp, callback) {
                    ___//process xmlhttp object into a result
                    ___callback(result);
                    }

                    function getsomething() {
                    ___myxhr('http://example.com', 'a=1', function(result) {
                    ______if (result.blah === 'gorp') {
                    _________dosomething;
                    ______} else {
                    _________dosomethingelse();
                    ______}
                    ___}
                    }

                    Now your getsomething function controls everything and all your xmlhttp
                    requests can be handled the same way without having to make a hardcoded
                    call. If you want the logic in the myxhrreturn function, you can put it
                    there, but it's not required. Here's another way to do it:

                    function getsomething() {
                    var cb;
                    if (somecondition) {
                    ______cb = function(result) {dosomething(result)};
                    ___} else {
                    ______cb = function(result) {dosomethingelse(result);};
                    ___}
                    ___myxhr('http://example.com', 'a=1', cb}
                    }

                    See how versatile this is? You only need a single callback, no extra Boolean
                    parameter is needed to indicate whether to call a hard-coded function name.
                    Any time you want more stuff to happen, you can load it into the passed
                    callback function somehow.

                    Erik

                    On Tue, Jan 4, 2011 at 1:47 AM, Felix E. Klee <felix.klee@...> wrote:

                    >
                    >
                    > On Tue, Jan 4, 2011 at 7:53 AM, Erik Eckhardt <erik@...<erik%40eckhardts.com>>
                    > wrote:
                    > > I thought you were using a Boolean to indicate whether to make the
                    > > return call to f.
                    >
                    > No, that was correct. In the example code, I *was* using a boolean. Only
                    > in the real code, things are more complex:
                    >
                    > Based on what the XHR returns some calculations are performed, and - if
                    > needed - another XHR is made.
                    >
                    >
                    > > The point I was trying to get across is that you don't need the
                    > > circularity.
                    >
                    > What's bad about the circularity?
                    >
                    >
                    > > xhr(params, function() {callback(); f();}); // circular call back to f
                    >
                    > That would again be an infinite loop. The body of the anonymous function
                    > would need to be at least slightly more complex. I don't think this is
                    > increases readability.
                    >
                    >


                    [Non-text portions of this message have been removed]
                  • Felix E. Klee
                    On Tue, Jan 4, 2011 at 7:35 PM, Erik Eckhardt ... There may be a misunderstanding. Thus, for your pleasure, below a more real-life
                    Message 9 of 17 , Jan 5, 2011
                    • 0 Attachment
                      On Tue, Jan 4, 2011 at 7:35 PM, Erik Eckhardt <erik@...>
                      wrote:
                      > no extra Boolean parameter is needed to indicate whether to call a
                      > hard-coded function name.

                      There may be a misunderstanding.

                      Thus, for your pleasure, below a more real-life example. Asides from
                      being more verbose, the only difference to my original example is the
                      addition of the parameter "liveCommentary". Together with the flag
                      "matchIsStillRunning" (formerly: "moreToDo") it forms the data returned
                      from the server.

                      var sendToServer; // defined elsewhere

                      function onCommentaryReceived(commentary, matchIsStillRunning) {
                      // write live commentary: ...
                      if (matchIsStillRunning) {
                      requestCommentary();
                      }
                      }

                      function requestCommentary() {
                      sendToServer('Berlin:Munich', onCommentaryReceived);
                      }

                      Naturally, one could rewrite this:

                      var sendToServer; // defined elsewhere

                      function requestCommentary() {
                      sendToServer('Berlin:Munich',
                      function (commentary, matchIsStillRunning) {
                      // write live commentary: ...
                      if (matchIsStillRunning) {
                      requestCommentary();
                      }
                      });
                      }

                      The second form avoids the circular function definition, but is it more
                      readable?
                    • Jordan
                      I think the confusion here is that you shouldn t be using the circular function pattern you are using. In addition, you are using function declarations
                      Message 10 of 17 , Jan 5, 2011
                      • 0 Attachment
                        I think the confusion here is that you shouldn't be using the "circular function" pattern you are using. In addition, you are using function declarations rather than assigning a function statement to a variable.

                        What you want to do, it seems, is immediately run requestCommentary after a matchIsStillRunning response comes back from the server.

                        What I can think of off the top of my head is:

                        var onCommentaryReceived = function (commentary, matchIsStillRunning) {
                        // write live commentary: ...
                        if (matchIsStillRunning) {
                        sendToServer('Berlin:Munich', onCommentaryReceived);
                        }
                        },
                        requestCommentary = function () {
                        return onCommentaryReceived(null, true);
                        };

                        Alternatively, and with less refactoring:
                        var onCommentaryReceived, requestCommentary;
                        onCommentaryReceived = function (commentary, matchIsStillRunning) {
                        // write live commentary: ...
                        if (matchIsStillRunning) {
                        requestCommentary();
                        }
                        };
                        requestCommentary = function () {
                        sendToServer('Berlin:Munich', onCommentaryReceived);
                        };

                        --- In jslint_com@yahoogroups.com, "Felix E. Klee" <felix.klee@...> wrote:
                        >
                        > On Tue, Jan 4, 2011 at 7:35 PM, Erik Eckhardt <erik@...>
                        > wrote:
                        > > no extra Boolean parameter is needed to indicate whether to call a
                        > > hard-coded function name.
                        >
                        > There may be a misunderstanding.
                        >
                        > Thus, for your pleasure, below a more real-life example. Asides from
                        > being more verbose, the only difference to my original example is the
                        > addition of the parameter "liveCommentary". Together with the flag
                        > "matchIsStillRunning" (formerly: "moreToDo") it forms the data returned
                        > from the server.
                        >
                        > var sendToServer; // defined elsewhere
                        >
                        > function onCommentaryReceived(commentary, matchIsStillRunning) {
                        > // write live commentary: ...
                        > if (matchIsStillRunning) {
                        > requestCommentary();
                        > }
                        > }
                        >
                        > function requestCommentary() {
                        > sendToServer('Berlin:Munich', onCommentaryReceived);
                        > }
                        >
                        > Naturally, one could rewrite this:
                        >
                        > var sendToServer; // defined elsewhere
                        >
                        > function requestCommentary() {
                        > sendToServer('Berlin:Munich',
                        > function (commentary, matchIsStillRunning) {
                        > // write live commentary: ...
                        > if (matchIsStillRunning) {
                        > requestCommentary();
                        > }
                        > });
                        > }
                        >
                        > The second form avoids the circular function definition, but is it more
                        > readable?
                        >
                      • Felix E. Klee
                        ... The question is: Why? Is that considered bad practice? If so, a reference please. ... See my original post. There I mentioned that I would like to enforce
                        Message 11 of 17 , Jan 6, 2011
                        • 0 Attachment
                          On Thu, Jan 6, 2011 at 6:27 AM, Jordan <ljharb@...> wrote:
                          > I think the confusion here is that you shouldn't be using the
                          > "circular function" pattern you are using.

                          The question is: Why? Is that considered bad practice? If so, a
                          reference please.

                          > In addition, you are using function declarations rather than assigning
                          > a function statement to a variable.

                          See my original post. There I mentioned that I would like to enforce
                          that convention for the whole project using JSLint. Only I wonder why
                          there is no such option.

                          > What I can think of off the top of my head is:
                          >
                          > [...]
                          > requestCommentary = function () {
                          > return onCommentaryReceived(null, true);
                          > };

                          That would work, but it's confusing to read. At least the function name
                          "onCommentaryReceived" should be changed.

                          > Alternatively, and with less refactoring:

                          See my original post. It's there already.
                        • Felix E. Klee
                          On Fri, Dec 31, 2010 at 6:33 PM, Felix E. Klee ... Just figured that this is akin to: http://en.wikipedia.org/wiki/Mutual_recursion And I
                          Message 12 of 17 , Jan 8, 2011
                          • 0 Attachment
                            On Fri, Dec 31, 2010 at 6:33 PM, Felix E. Klee <felix.klee@...>
                            wrote:
                            > If I write code such as the following

                            Just figured that this is akin to:

                            http://en.wikipedia.org/wiki/Mutual_recursion

                            And I checked: The latest version of JSLint, 2011-01-06, still
                            complains about an undefined function.
                          • Douglas Crockford
                            ... ` var sendToServer; ` ` function f() { ` sendToServer( xyz , function (moreToDo) { ` // do something ... ` if (moreToDo) { `
                            Message 13 of 17 , Jan 8, 2011
                            • 0 Attachment
                              --- In jslint_com@yahoogroups.com, "Felix E. Klee" <felix.klee@...> wrote:
                              >
                              > If I write code such as the following, then JSLint complains that "f" is
                              > not defined.
                              >
                              > var sendToServer;
                              >
                              > function callback(moreToDo) {
                              > // do something ...
                              > if (moreToDo) {
                              > f();
                              > }
                              > }
                              >
                              > function f() {
                              > sendToServer('xyz', callback);
                              > }
                              >
                              > What's the most elegant solution to get rid of the error message?


                              ` var sendToServer;
                              `
                              ` function f() {
                              ` sendToServer('xyz', function (moreToDo) {
                              ` // do something ...
                              ` if (moreToDo) {
                              ` f();
                              ` }
                              ` });
                              ` }
                            • Felix E. Klee
                              On Sat, Jan 8, 2011 at 4:05 PM, Douglas Crockford ... Thanks for the suggestion! I guess I ll take this approach and put the lengthy do something into a
                              Message 14 of 17 , Jan 8, 2011
                              • 0 Attachment
                                On Sat, Jan 8, 2011 at 4:05 PM, Douglas Crockford
                                <douglas@...> wrote:
                                > ` var sendToServer;
                                > `
                                > ` function f() {
                                > ` sendToServer('xyz', function (moreToDo) {
                                > ` // do something ...
                                > ` if (moreToDo) {
                                > ` f();
                                > ` }
                                > ` });
                                > ` }

                                Thanks for the suggestion! I guess I'll take this approach and put the
                                lengthy "do something" into a separate function.

                                Just thinking about it: I assume that having JSLint to *not* report an
                                error on circular function definitions is non-trivial. That is, if one
                                still wants to be warned about cases were an undefined function would be
                                called, such as:

                                function f() {
                                g();
                                }

                                f();

                                function g() {
                                // do something
                                }
                              Your message has been successfully submitted and would be delivered to recipients shortly.