Ninja Pumpkin Mutants


    1.3.5 • Public • Published


    a browser-based loader that works with both AMD and CommonJS modules (with some caveats)


      • DO NOT USE
    • check it out

    • Allows for on-demand requires in AMD modules

    • Allows imports of CommonJS modules into browser

      • CAVEAT: cjs module can only have top-level requires. Nested requires (i.e. those within a function) will not work (because that function would need to be made async, then any function using that function would also need to be made async, and so on...)

    Some alternatives (though not complete as per our reqs):

    Read NPM package.json DOC and also this bit about the .browser field

    • es2015 === es6 modules: import x from 'y';
    • commonjs2: for node, module.exports = entry_return; // uses require('dep') for dependencies
    • amd: for browsers, define([...deps],function(...deps){})
      • should not use require; all deps are upfront in deps array


    • define(name, [deps], fcn) ===> modules[name/url] = fcn(...resolvedDeps);

      • this is how to define a module
      • fcn executed once, then cached for that name
        • fcn must RETURN its module (e.g. an object of methods)
      • fcn executed AFTER all deps are loaded & resolved
    • require([deps], fcn) is how to USE/EXECUTE a module:

      • module is represented by fcn
      • module is executed by calling fcn(deps)
        • but only once all deps are loaded
    • todo: could allow require from within an AMD module (use same technique as for commonjs modules)

      • not standard, but so what... (simplifies writing manually-written plugins perhaps?)
    // TO DOCUMENT: 
    // - can define GLOBALS to better "sandbox" module initialization
    //     - can 'protect' framework code: e.g. redefine window. or console. or alert/confirm
    // - custom url resolvers
    // - custom loaders (css, json, ...)
    //     - can load modules other than just javascript (text, json, css, objects, data)
    // once loaded, modules are reused as initially loaded
    // BUT, what happens if module is loaded with custom settings?
    // - should that module be loadable again with different settings?
    // - may want to reload an existing module but with different parameters (i.e. config)
    // - save source? retry if config now different then config then?
    /* generally background info:
        important fyi:
        - dynamic import() is NOT SUPPORTED by most modern browsers:
           - as of jan 21, 2019: edge=no, firefox=no, chrome=yes, safari=yes
    /* loadModule: similarities and differences from AMD's define structure:
        - no last param as function: 
            - use .then([...loadedModules]) instead
                - loadModule returns a promise that completes after all modules are loaded
        - loadModule NEVER FAILS: 
            - 'await loadModule' ALWAYS succeeds: 
                - no need to try/catch; catch clause will never execute
                - only .then clause needed; .catch clause will ever be called
            - problem modules (e.g. network|syntax|initialization errors) are resolved
              as the Error that caused the load failure
                - dependents can test for 'mod instanceof Error'
            - individual modules WILL be resolved as:
                - undefined: if nothing is returned from their execution
                    - module loading is for its side-effects (e.g. sets up global vars, like jQuery, lodash)
                - Error: if loading generates an error (network|download|syntax|initialization error)
                - actual module if all is well
                    - e.g. axios
        - loadModule can be called with 1 or more parameter:
            - each params treated as individual modules
                - can load a single module: 
                        const mod = loadModule('mod-url');
                - can load multiple modules (using array destructuring): 
                        const [mod1, mod2, mod3] = await loadModule('mod1-url', 'mod2-url', 'mod3-url');
            - each module is loaded ONLY ONCE
                - ID for that module is its resolved url
                - any further request for that module returns original initialized module
            - return for 'await loadModule(...)' is:
                - single module, if using single param
                    - explicit: 
                            const mod = await loadModule('mod-url');
                    - then clause: 
                              .then(mod => {...code using module...});
                - [...array of modules...], if using 2 or more params
                    - array of modules, can be destructured as follows:
                        - explicit: 
                              const [mod1, mod2, mod3] = await loadModule('mod1-url', 'mod2-url', 'mod3-url');
                        - then clause: 
                              loadModule('mod1-url', 'mod2-url', 'mod3-url')
                                .then(([mod1, mod2, mod3]) => {...code using modules...});
                - if don't know how many modules loaded (e.g. from a ...computedArray), 
                        .then(modOrMods => {
                            // resolvedMods (below) is ALWAYS an array
                            const resolvedMods = Array.isArray(modOrMods) ? modOrMods : [modOrMods];
                            // alternative:
                            // Array.isArray(modOrMods) || (modOrMods = [modOrMods]); 
                            // alternative:
                            const mods = Array([modOrMods]).flat(); // but array.flat() is NOT supported by Edge
        //  - arg can be a string or not
        //      - if it's a string, it's loaded as per below
        //      - if it's not a string, the module's value is that arg
        //          - only sensical purpose for this is there is an onReady function as last param
        // if only 1 arg, that module is returned
        // if 2 or more args, loaded modules are returned as an array
        [read this](
        [also this for jsdelivr](
        [Some notes from here](
        - in package.json:
            - main is for UMDs
            - module is for ES6/import-export
            - also seen in the wild: 
                - browser (presumably amd)
                - unpkg: guess... :-)
        - QUESTION: should we ALWAYS return an array instead?
            - then there's no guessing
                - always destructure it
                    const [mod] = loadModule('single-mod-url');
                        .then(([mod]) => {...code using single module...});
                    const [mod1, mod2, mod3] = await loadModule('mod1-url', 'mod2-url', 'mod3-url');
                    loadModule('mod1-url', 'mod2-url', 'mod3-url')
                        .then(([mod1, mod2, mod3]) => {...code using modules...});
                - use array[0] if know it's a single mod
            - maybe we export loadModules() and loadSingleModule()?
                - but then, which is default?
            - SO, question is, how likely/often will we load modules WITHOUT knowing specifically which 
              or how many modules we're loading... (i suspect, not many)
    /* loadModule parameters:
        - each param can be:
            - string (i.e. typeof param === 'string')
            - not a string (i.e. typeof param !== 'string')
        - if param is NOT a string
            - module becomes that object
                - no further processing occurs
        - if param is a string:
            - format is: [name=] [type: | type!] dataOrUrl
            - where:
                name: used to assign resulting module (as loaded & initialized)
                as window[name] = moduleValue;
                type: format is loaderName | loaderName-data | data
                        - if data or -data present, rest of string (including multilines)
                          is treated as explicit immediate data
                        - if no [-]data present, rest of string (presumably single line; meaningless otherwise)
                          is treated as a url to be resolved later
                dataOrUrl: either:
                - data (as specified by type)
                - a url, to be resolved by url resolvers
            All module content (immediate data or downloaded from a url) is then loaded
            through a loader
        - url resolving:
            - if starts with 'https://' or '//',
                - url is used as-is
            - if starts with '.' or '/'
                - url is used as relative to main html page (i.e. window,location.href)
                    or url of parent module that requested it (e.g.
            - anything else is treated as an NPM package and downloaded from:
            custom resolvers:
                - can be added via config and will take precendence over default resolvers (above)
                    - to override a resolver, use a matching scheme that covers the url-scheme/format
        - loaders:
            - once downloaded (or if immediate data), content is processed through
                applicable loader (by type if explicit, by content-type if downloaded, 
                or by its extension if content-type not available)
            - each loader can pre-process content as needed, then returned [possibly] updated
                content, if needed (e.g. turn json content into an actual object using JSON.parse)
            - each loader can take actions on behalf of the content (e.g. load css code directly
                into the html page)
            - if [content-]type is javascript, code will THEN (i.e. after pre-processed by loader)
                be executed/initialized
            default (builtin) loaders:
            - css
                - loads css into html page
                    - returned module value is css code
            - json
                - converts json content into an object
            - javascript
                - to prevent auto-initialization, use 'text' or 'data' as type
                    - then will be left alone
                - can specify custom type
                    - then pre-process and/or load as needed
            - custom loaders:
                - can be added via config and will take precendence over default loaders
                    - to override a loader, use a matching scheme that covers the content type
    /* examples of module string references:
            download from jsdelivr and load (i.e. initialize) module
            download from jsdelivr and load (i.e. initialize) module; then assign that module to window.axs
            download from jsdelivr but treat as text (so do NOT initialize); assign downloaded content (presumably text) to window.axs
        name=url // download and loads as css or json-param
        name=//url [same as above]
        name=data:...plain text data...
        name=text-data:... load and assign (same as above)
        name=json-data:... load and assign as object
        css:url // loaded
        json:url // param
        text:url // param always (even if text or css)
        'url' // downloaded as abs; then, if css-loaded; if json, kept for param (need function); if JS, kept for exec later
        'name=url', // assign to windows download to windows; if css, loaded first
        'name=raw!data' // assign text data as name
        'name=raw-json!data' // load as json, assign to name
        'json!url' // usable only if a function is specified else no purpose
        http-data:??? [download as text?]
        https-data:??? [download as text?]
        text:url [download and use as-is]
        raw:url [download and use as-is]
        data:... [same as text-data; do nothing with data]
        css-data:... [loads as style]
        json-data:... [convert to obj]
        - '[name=]string...' -> loads from unpkg
        - '[name=]./string...' -> loads relative to base url
        - '[name=]http://...', '[name=]https://...', '[name=]//...' loads directly from absolute url (also becomes base url for sub-dependencies)
        - object { main, css, json, text, base } // script? loaded as script instead of function
            - main: js code (downloaded first, executed last)
                - if array, executed left-to-right
            - css: downloaded first, loaded into page/doc
            - json/text: downloaded next
            - data/img: ???
        - if [name=], result (or data) assigned to
        - if last param is function(dep1,dep2,...)
            - num dep parms must match num prior args
            - each prior arg's value becomes matching dep
            - function executed last (same as an AMD define call)
        - if download content-type or url extension is:
            - application/javascript, will be executed
            - /css: will be loaded as style
            - all others downloaded and used as parmDep,, or kept as its module name for later use
                - /json: will be loaded as data (then used as parmDep,, or kept as its module name for later use)
                - text/: try to decode? or keep as text data?
                - binary: ? image? stream? (video, audio? pdf? other) how to keep, how to use
        - AMD define depends on module being pre-bundled/packaged
        - our loadModule allows for "manual define" to be done on-the-fly
    // MAJOR ISSUES with UNPKG CDN: was broken for a few days before/during Feb 4, 2019
    // (and a few times in prior months/years for a variety of reasons: bugs, upgrades, ...)
    // SO, we decided to use CDN instead
    // - documentation:
    // one benefit: automatically serves content WITHOUT redirects even when not specifying exact versions (i.e. server-side redirection)
    // another: add .min to cs/js files and will ALWAYS serve minified (if not already minified, jsdelivr will minify for you: may be slower initially)
    /* another alternative:
        - uses yet another CDN:
        - format seems to be:[LIB_NAME_HERE]/[VERSION.HERE]/[FILE-NAME.HERE]
        - e.g.
        // based on:
        // const cdnjs = what => `${what}`,
        //       search = name => cdnjs(`libraries?search=${name}`), // .results = [], .total = number; each result: { .name, .latest: file-url-of-latest-version (presumably minifiled)}
        //       lib = name => cdnjs(`libraries/${name}`), // .name, .filename, .assets = [{ .version, .files=[ 'core.js', 'jquery.js', 'jquery.min.js',... ]}...]
        //       libv = name => cdnjs(`libraries/${name}?fields=name,filename,version`);
    // Using WITH and PROXY to create an execution context (e.g. for a plugin):
    // -
    // - read:
    // VERY convenient to add GLOBAL variables without explicitly defining it them upfront
    // as required when using new Function(); good when there are MANY such variables
    // BUT, gist above is NOT secure:
    //  can ALWAYS access GLOBAL context: (0,eval)('this')
    // UNLESS override EVAL && FUNCTION directly
    // because if not ovveriden, can always access it directly using:
    //    Object.getPrototypeFor(function(){}).constructor(...stringParms, 'function code here');
    ## Security
    // BEST APPROACH to security for now: warn that dynamically loaded code has ACCESS TO EVERYTHING
       Trying to sandbox dynamic code by protecting/securing the global window object 
       - it can add a lot of code (as per below)
       - a large amount of overhead (so likely performance hit)
       - and unless each case is considered individually, it's not clear that dynamic code
         will be appropriately sand-boxed
       - can always load by adding script tag (or iframe or img or audio/video or ?) and that would
         be VERY hard to prevent (unless prevent document manipulation)
       code below was an attempt in that direction: 
       - it is NOT currently used anywhere (or tested, or completed)
       - it MUST NOT be used anywhere else
       - it is kep here strictly for reference
        ### Javascript REALMS WILL SOLVE THESE issues
        - we'll wait for this to become part of the language
        - currently at stage 2 in TC39
        There is no easy way to prevent module code from accessing main window object, 
        even if proxying all vars using 'with' because even without using any names, 
        Object.getPrototypeOf(function(){}).constructor(...parms, `window; (0,eval)('this')`) will
        gain access to top window unless CHANGE (function(){}).constructor itself
        which would change it FOR EVERYONE (which may be OK)
        - still does NOT prevent code from explicitly loading directly from script element
        ### REALMS are likely what I'm looking for to execute 3rd-party code using a separate 
        // (i.e. self-contained) global context;
        // TODO: wait until REALMS/SES are at TC39 Stage 3 level before implementing
        //       - in the meantime, NO global security for dynamic modules/plugins
        //       - BUT:
        //          - no worse than, say, jquery plugins, that work in main/global/user space anyway
        //          - no worse than php wordpress plugins that also work in user space and can take down
        //            a site, or spy for information
        // - READ about REALMS and SES:
        //   - read: (forum about javascript items, esp. relating to security)
        //   - REALMS:
        //      - not currently transpilable from babel (feb 12, 2019)
        //      - best to wait until realms at least at stage 3
        //      - read:
        //      - read:
        //      - read:
        //   - SES: Secure EcmaScript
        //      -
        //      - based on Google CAJA SES
        //        -
        //        -

    Issues when loading AMD and CJS modules

    Unless we know ahead of time if a module is AMD or CJS, you have 2 possible strategies to load them: 1- pass both (define,require) and (exports, module.exports) then see which one was used after initialization 2- pass 1 (e.g. define, require) and if that fails pass the other (in this case exports, module.exports)

    The problem with 2 passes is that at least some of thecode will execute twice and this may be bad (e.g. if some one time init must take place, so doing it twice will corrupt/affect/change data)

    The problem with (1) is that in order to do it right, the module's code must be scanned every time in order to extract the require's name (i.e. the dependencies) in order to preload them before running the init code.

    Problems with above strategy:

    1- preloading will only work if the deps are static (i.e. single string); if the dep is an expression (so likely dynamic) there is no way to preload it (no way to extract it); - we could modify the code to add an 'await' to then allow for resolution at runtime, with 2 more caveats: 1- unless the await is at the top-level (where we can control loading as async), a nested require will likely introduce a syntax error (because not in an async function); and it's a fool's errand to try to 'async-tify' these methods, then those calling them, and so on... 2- if the require is itself part of a longer expression (e.g. require('x').prop or require('x')(, the simple await will break the call because await is right associative meaning it will now await the whole expression (i.e. await (require('x').prop)) rather than the intended correction (i.e. (await require('x')).prop) 2- preloading assumes that those extracted dependencies are not already loaded, as would be the case with bundled modules (e.g. using webpack).

    In fact, although the requires might still be present in the source code, the bundler (e.g. webpack) expects to do its own loading of these modules from bundled code. If we now "extract" those dependencies then try to download them, we'd be doing this with likely invalid links (pointing to sources that only were relevant during the build period) and downloading code that does not need to be downloaded anyway (for example, a simple vue component can include 200 such requires, none of which need downloading, and most of which would return 404 if attempted)

    3- MIGHT be present because if optimized for production, the requires may also have been removed (e.g. webpack removes that part of the code) so when loading that code, extracting requires would not be a problem (since they have been removed by webpack) 2b- a more difficult issus still is if some of those bundled requires are not in fact embedded, in which case we'd need to know, somehow, to load those but not the others


    • when loading module, we always pass all four bits (define, require, exports, module.exports)
    • we always scan for requires UNLESS EXPLICITLY told not to
      • implies that you know whether or not a module is bundled or not

    so which is default???

    • if AMD, no requires but can preload dep-array

    • if AMD+simplified-cjs

      • scan for requires to preload as dep-array
    • if CJS, scan for requires to preload as dep-array

      • and everything below? only if relative?
    • if BUNDLED, no prescan: assumes deps are part of package

    • build array based on name of format for module

      • if 'cjs:axios' all axios now cjs
    • module names:

      • 'cjs:name'
        • EXTRACTS: expects cjs and extracts requires
      • 'amd:name' or 'umd:name' expects amd/umd and:
        • extract requires only if using simplified-commonjs format
        • does not extract requires otherwise
          • but does preload [array-based-dependencies]
      • 'bundled' does not extract requires (assumes part of bundle)
        • can be either cjs or amd/umd
        • so acts as self-contained OR
      • scan for requires or not:
        • preload
        • no-preload
        • no-pre-scan (default is scan)
        • bundled (no scan); can be either cjs or amd/umd
        • cjs: scan for requires always
          • unbundled: same but might be amd/umd
        • unbundled (scan & preload requires)


    npm i load-dynamic-module

    DownloadsWeekly Downloads






    Unpacked Size

    319 kB

    Total Files


    Last publish


    • frudman