Leaderboard

In this tutorial I want to write about handling special cases and change requests by clients gracefully without introducing code bloat or degrading code quality and maintainability. I'll use a site's navigation menu as an example, as it's relatable and pretty much every site has one. I'll give some examples of real situations and change requests I encountered during projects, and describe multiple approaches to handling them. However, this post is also about the general mindset I find useful for ProcessWire development, which is more about how to handle special cases and still keep your code clean by making the special case a normal one. The problem: Special cases everywhere Since ProcessWire has a hierarchical page tree by default, as a developer you'll usually write a function or loop that iterates over all children of the homepage and displays a list of titles with links. If the site is a bit more complex, maybe you additionally loop over all grandchildren and display those in drop-down menus as well, or you even use a recursive function to iterate over an arbitrary amount of nested child pages. Something like this: function buildRecursiveMenu(Page $root): string { $markup = ['<ul class="navigation">']; foreach ($root->children() as $child) { $link = '<a class="navigation__link" href="' . $child->url() . '">' . $child->title . '</a>'; $children = $child->hasChildren() ? buildRecursiveMenu($child) : ''; $markup[] = "<li class="navigation__item">{$link}{$children}</li>"; } $markup[] = '</ul>'; return implode(PHP_EOL, $markup); } But then the requests for special cases come rolling in. For example, those are some of the requests I've gotten from clients on my projects (by the way, I'm not saying the clients were wrong or unreasonable in any of those cases - it's simply something I needed to handle in a sensible way): The homepage has the company's name as it's title, but the menu link in the navigation should just say "Home". The first page in a drop-down menu should be the top-level page containing the drop-down menu. This was requested because the first click on the top-level item opens the sub-navigation instead of navigating to that page (espcially on touch devices, such as iPads, where you don't have a hover state!), so some visitors might not realize it's a page itself. Some top-level pages should be displayed in a drop-down menu of another top-level page, but the position in the page tree can't change because of the template family settings. The menu needs to contain some special links to external URLs. For one especially long drop-down menu, the items should be sorted into categories with subheadings based on a taxonomy field. In general, my solutions to those requests fall into three categories, which I'll try to elaborate on, including their respective benefits and downsides: Checking for the special case / condition in the code and changing the output accordingly (usually with hard-coded values). Separating the navigation menu from the page tree completely and building a custom solution. Utilizing the Content Management Framework by adding fields, templates and pages that represent special states or settings. Handling it in the code This is the simplest solution, and often the first thing that comes to mind. For example, the first request (listing the homepage as "Home" instead of it's title in the navigation) can be solved by simply checking the template or ID of the current page inside the menu builder function, and changing the output accordingly: // ... $title = $child->template->name === 'home' ? 'Home' : $child->title; $link = '<a class="navigation__link" href="' . $child->url() . '">' . $title . '</a>'; // ... This is definitely the fastest solution. However, there are multiple downsides. Most notably, it's harder to maintain, as each of those special cases increases the complexity of the menu builder function, and makes it harder to change. As you add more special conditions, it becomes exponentially harder to keep changing it. This is the breeding ground for bugs. And it's much harder to read, so it takes longer for another developer to pick up where you left (or, as is often cited, for yourself in six months). Also, now we have a hard-coded value inside the template, that only someone with access to and knowledge of the template files can change. If the client want's the link to say "Homepage" instead of "Home" at some point, they won't be able to change it without the developer. Also, each special case that is hidden in the code makes it harder for the client to understand what's going on in terms of template logic - thus increasing your workload in editorial support. That said, there are definitely some times where I would go with this approach. Specifically: For smaller projects that you know won't need to scale or be maintained long-term. If you are the only developer, and/or only developers will edit the site, with no "non-technical" folk involved. For rapid prototyping ("We'll change it later") Building a custom solution My initial assumption was that the main navigation is generated based on the page tree inside ProcessWire. But of course this isn't set in stone. You can just as easily forgo using the page tree hierarchy at all, and instead build a custom menu system. For example, you could add a nested repeater where you can add pages or links on a general settings page, and generate the menu based on that. There are also modules for this approach, such as the Menu Builder by @kongondo. This approach is not the quickest, but gives the most power to the editors of your site. They have full control over which pages to show and where. However, with great power comes great responsibility, as now each change to the menu must be performed manually. For example, when a new page is added, it won't be visible in the menu automatically. This is very likely to create a disconnect between the page tree and the menu (which may be what you want, after all). You may get ghost pages that are not accessible from the homepage at all, or the client may forgot to unpublish pages they don't want to have any more after they've removed them from the menu. I would only go with this approach if there are so many special cases that there hardly is a "normal case". However, even then it might not be the best solution. The direct relationship between the page tree, the menu structure and page paths are one of the strongest features of ProcessWire in my opinion. If many pages need to be placed in special locations without much structure in regards to what templates go where, maybe you only need to loosen up the template family settings. I have built one site without any template family restrictions at all - any page of any template can go anywhere. It's definitely a different mindset, but in this case it worked well, because it allowed the client to build custom sections with different page types grouped together. It's a trade-off, as it is so often, between flexibility and workload. Weigh those options carefully before you choose this solution! Utilizing the CMF This is the middle ground between the two options above. Instead of building a completely custom solution, you keep with the basic idea of generating a hierarchical menu based on the page tree, but add fields and templates that allow the editor to adjust how and where individual pages are displayed, or to add custom content to the menu. of course, you will still write some additional code, but instead of having hard-coded values or conditions in the template, you expose those to the client, thereby making the special case one of the normal cases. The resulting code is often more resilient to changing requirements, as it can not one handle that specific case that the client requested, but also every future change request of the same type. The key is to add fields that enable the client to overwrite the default behaviour, while still having sensible defaults that don't require special attention from the editor in most cases. I'll give some more examples for this one, as I think it's usually the best option. Example 1: Menu display options This is probably the first thing you thought of for the very first change request I mentioned (displaying the homepage with a different title). Instead of hard-coding the title "Home" in the template, you add a field menu_title that will overwrite the normal title, if set. This is definitely cleaner than the hard-coded value, since it allows the client to overwrite the title of any page in the menu. I'll only say this much in terms of downsides: Maybe the menu title isn't really what the client wanted - instead, perhaps they feel limited because the title is also displayed as the headline (h1) of the page. In this case, the sensible solution would be an additional headline field that will overwrite the h1, instead of the menu_title field. Which fields are really needed is an important consideration, because you don't want to end up with too many. If each page has fields for the title, a headline, a menu title and an SEO-title, it's much more complicated than it needs to be, and you will have a hard time explaining to the client what each field is used for. Another example in this category would be an option to "Hide this page in the menu". This could be accomplished by hiding the page using the inbuilt "hidden" status as well, but if it's hidden it won't show up in other listings as well, so separating the menu display from the hidden status might be a good idea if your site has lots of page listings. Example 2: "Menu link" template One solution that is quite flexible in allowing for custom links to pages or external URLs is creating a menu-link template that can be placed anywhere in the page tree. This templates can have fields for the menu title, target page and/or external target URL. This way, you can link to another top-level page or an external service inside a drop-down menu, by placing a Menu Link page at the appropriate position. This is also a clean solution, because the navigation menu will still reflect the page tree, making the custom links visible and easily editable by the editors. A minor downside is that those templates are non-semantical in the sense that they aren't pages with content of their own. You'll need to make sure not to display them in listings or in other places, as they aren't viewable. It may also require loosening up strict family rules - for example, allowing for Menu Link pages to be placed below the news index page, which normally can only hold news pages. Example 3: Drop-down menu override This one is a more radical solution to override drop-down menus. You add a repeater field to top-level pages, similar to the one mentioned as a custom solution, where you can add multiple links to internal pages or URLs. If the repeater is empty, the drop-down menu is generated normally, based on the sub-pages in the page tree. But if the repeater contains some links, it completely overrides the drop-down menu. It's similar to the fully custom solution in that as soon as you override a sub-menu for a top-level page, you have to manually manage it in case the page structure changes. But you can make that decision for each top-level page individually, so you can leave some of them as is and only have to worry about the ones that you have overwritten. Again, this offers sensible defaults with good customizability. A downside is that the mixed approach may confuse the client, if some changes to the page tree are reflected in the drop-down menu directly, while others don't seem to have any effect (especially if you have multiple editors working on a site). Finding the right solution So how do you choose between the approaches? It depends on the client, the requirements, and on what special cases you expect and want to handle. Sometimes, a special request can be turned down by explaining how it would complicate editorial workflows or have a negative impact on SEO (for example, if you risk having some pages not accessible from the homepage at all). Also, make sure you understand the actual reason behind a change request, instead of just blindly implementing the suggestion by the client. Often, clients will suggest solutions without telling you what the actual problem is they're trying to solve. For example: In one case, I implemented the drop-down override mentioned in example three. However, what the client really wanted was to have the top-level page as the first item in the drop-down menu (see the example requests mentioned above). So they ended up overwriting every single drop-down menu, making the menu harder to maintain. In this case, it would have been better to go with a more specialized solution, such as adding a checkbox option, or even handling it in the code, since it would have been consistent throughout the menu. Another example was mentioned above: If the client requests an additional "Menu title" field, maybe what they really need is a "Headline" field. I recommend reading Articulating Design Decisions by Tom Greever; it includes some chapters on listening to the client, finding out the real reason behind a change request, and responding appropriately. It's written from a design perspective, but is applicable to development as well, and since UX becomes more important by the day, the lines between the disciplines are blurred anyway. Conclusion I realize now this reads more like a podcast (or worse, a rant) than an actual tutorial, but hopefully I got my point across. ProcessWire is at is greatest if you utilize it as a Content Management Framework, creating options and interfaces that allow for customizability while retaining usability for the client / editor. I usually try to hit a sweet spot where the editors have maximum control over the relevant aspects of their site, while requiring minimal work on their part by providing sensible defaults. Above, I listed some examples of requests I've gotten and different solutions I came up with to handle those with custom fields or templates. Though in some cases the requirements call for a custom solution or a quick hack in the template code as well! What are some of the special requests you got? How did you solve them? I'd love to get some insights and examples from you. Thanks for reading!

Pushed a fix for RockMarkup2 and related modules (RockTabulator) not working in subdir installations. Thx for finding that @dragan

@ryan No, not yet. I have only enabled a few functions. What I found strange though (perhaps just coincidence): The bottom left link "disable AoS" also stopped working as it should, i.e. I had to save module settings twice, before it was disabled. That link to disable the module has a query string parameter toggle, which is set to 1. And afaik the first toogle field-value (e.g. Yes) also has value 1. That's maybe one toggle too many :-)

@Robin S Not yet interactively, though it is possible with hooks. Though I do plan to make it part of the "Form" field settings once the field gets more mature. Yes. Existing dependencies in forms you integrate will continue to work, and you can also make fields in the main form be dependent upon fields within an integrated form. The field name for fields in an integrated form is "formName_fieldName" rather than just "fieldName". That's how you can integrate multiple copies of the same form, and still end up with unique field names. So if using dependencies, then you just refer to the "formName_fieldName" in your dependency. For existing dependencies in an integrated form, FormBuilder takes care of converting those for you, so that the dependency works regardless of where else the form is used. Each page is validated independently. It creates a partial entry on the server side (in the DB) and saves it on submit of every pagination. If there are errors on the pagination you submit, it will stay on that pagination until you fix them. Yes. Partial entries also have their own URL (unique query string) that can be bookmarked or returned to if someone doesn't want to complete it all in one setting or their session expires, etc. Yes, but probably not in the first version. Though this is part of the reason why it's setup so that partial entries have their own URL. The page number can be specified in the query string. Any pagination has access to the entire entry server side (should you want to examine anything with hooks), but the front-end/client side only knows about the current pagination. So dependencies across paginations won't work at present. If there's interest in the feature, the plan was to support it by rendering the non-present Inputfields referred to on any pagination's showIf/requiredIf values within the current pagination, but behind a hidden element. In that manner, they are technically present for the pagination's processing and front-end JS, even if not visibly present.

Yep, I was also underwhelmed, with both services. An upscaled image (4x) showed really ugly artefacts on the face and hands parts of an image. With background-removal... well, an entire (huge) part between two body-parts was not removed (similar skin-tone and background-color). It seems that even with AI in 2019, the old adage still holds true: SISO. The better / bigger your original image is, the better output you'll get.

I'm not that impressed. letsenhance.io vs Photoshop: Original: downscaled to upscale ? 4x upscale letsenhance.io – it took about 10 minutes! 4x upscale Photoshop - under 1 second: letsenhance.io is not better just different is some ways. Certianly was not worth the wait of about 10 minutes, especially as its algorithm tends to distort geometric lines. We can argue that letsenhance.io is better at photos, but Photoshop is good at those too... Looks like letsenhance.io is only impressive at painting like squiggles when its own added squiggles fit the bill, maybe?

Putting "use strict" should really be the default... It simply shows more hints and errors and this can be very useful while debugging. Using this in every JS file now revealed some data type mismatch too, when using !=== instead of !==. (see screenshots below) In trash.js I changed basically these lines: var gridInstance = $('.RockTabulator.tabulator'); var grid = RockTabulator.getGrid(gridInstance); and further below: // success if(result.success) { UIkit.notification(check + result.success, {timeout: 3000}); if( RockTabulator.getGrid(gridInstance).reload() ) { console.log('grid has reloaded!'); } else { console.log('grid has not reloaded!'); } // grid.reload(); } }, But as I said, since RT is not known inside that JS script, it can't call an unknown method getGrid...

This doesn't warrant a full tutorial, but I wrote a little function that will recursively search through a repeater field to find the first non-empty field matching a list of fields to look for. I needed something like this to generate a fallback for SEO fields (og:image, og:description et c.). Here it is: <?php namespace ProcessWire; /** * Find the first non-empty field out of a list of fields in a repeater or repeater matrix field. * Will recursively search through nested repeaters until it finds a non-empty * field. * * @param RepeaterPageArray $repeater The field to search through. * @param array $fields A list of fields the function should look for. * @param array|null $allowed_repeater_types All the Repeater Matrix types the function will check. Leave empty to allow all. * @param array|null $allowed_repeater_fields All the repeater fields the function will check recursively. Leave empty to allow all. * @return void */ function firstRecursiveRepeaterMatch( RepeaterPageArray $repeater, array $fields, ?array $allowed_repeater_types = null, ?array $allowed_repeater_fields = null ) { // iterate over the items of the repeater foreach ($repeater as $current) { // if the function is currently inside a repeater matrix field, // skip this item if it isn't one of the allowed types, unless // allowed_repeater_types is empty (all types allowed) if ( $current instanceof RepeaterMatrixPage && is_array($allowed_repeater_types) && !in_array($current->type, $allowed_repeater_types) ) { continue; } // get all fields of the current item foreach ($current->getFields() as $field) { $name = $field->name; // if the current field is another repeater, check it recursively $fieldtype_class = $field->getFieldType()->className(); if ($fieldtype_class === 'FieldtypeRepeater' || $fieldtype_class === 'FieldtypeRepeaterMatrix') { // continue with the next item if the field name isn't one of the // allowed repeater fields, unless allowed_repeater_fields empty (null) if ( is_array($allowed_repeater_fields) && !in_array($name, $allowed_repeater_fields) ) { continue; } $deep_search = firstRecursiveRepeaterMatch( $current->get($name), $fields, $allowed_repeater_types, $allowed_repeater_fields ); // if the deep search inside the repeater // finds something, the function ends here if ($deep_search !== null) { return $deep_search; } } // if the current field name is one of the requested // fields, return it's value if it isn't empty if (in_array($name, $fields)) { $value = $current->get($name); if ( // check for empty values !empty($value) // if the value is any wirearray, check // if it has at least one item && (!$value instanceof WireArray || $value->count() > 0) ) { return $value; } } } } // if the function reaches this point, there is no match in the entire tree return null; } Can be used like this: $seo_description = firstRecursiveRepeaterMatch( // content sections fields, a repeater matrix with multiple types $page->sections, // look for the first non-empty instance of any of those fields ['body', 'html_basic', 'html_full'], // only check sections of the following types ['section_text', 'section_columns', 'section_accordion'], // only check the following nested repeaters recursively ['columns', 'accordion'] ); Wanted to share because I thought it could be useful to others. It should be easy to adjust the matching condition from non-emtpy fields to some other condition, depending on the use case ...

@szabesz Fair enough! ? @adrian As a user, I agree - i don't like it when I try to navigate away from a page using a link, and it opens in a new page, forcing me to go back to close the tab. But we're power users, I'm not sure everyone knows how to open links in a new tab (especially on mobile, not everyone gets the long-press interaction!). As for why I still use target blank basically for all external links, you know, business reasons. Almost every client sees red when they encounter a link that leads them away from their site. Not sure why everyone expects their audience to have the attention span of a mouse, but oh well ... it's not something I want to spend energy discussing every time. I can see both sides though.

@adrian @horst @szabesz Thanks! Interesting how those solutions are different yet similar. I usually do something like this for footer menus that tend to be more custom, since it only contains a cherry-picked selection of the most important pages. I have tried a couple different combinations of fields, with different levels of control. Since we're doing screenshots, here's the latest iteration that I find quite flexible: Sorry it's in German. Basically, there's a repeater "Footer regions", each containing a headline and a repeater with links. The regions get displayed as individual columns in the footer. Each link has a radio selection between "Internal page", "External URL" and "Download", and a "Link-Text" field. Depending on the radio selection, the respective field get displayed (page reference, URL or file Upload). If the Link-Text is empty, the link is displayed with a reasonable default (Page title for internal pages, domain for external URLs, file basename for files). I initially disliked having too use an additional field for the link-type, but it provides better usability for the client in my experience. @horst I like the solution with the show_only_if fields, though it might not be clear to clients how to switch it back? I found that the interaction with fields being only visible when other fields are empty/filled is not so intuitive for clients. @szabesz Interesting, is the "Open blank" option something your clients actively use? I usually just handle that in the template, i.e. all URLs to different hosts get target="_blank" rel="noopener" automatically ...

I also use different ways for implementing menus, depending on what I find appropriate. Here is a repeater based, one level customizable menu, where hiding fields on conditions makes it possible to switch between "custom" and page picker options:

Do you have your test files in subdirectories? The breadcrumb should show the subdirectory names (paths). On configurable I meant the "Show breadcrumbs" checkbox above the test list.

Not currently, but it could be a simple addition.

You can get UIKit purged down to ~25 - 40KB.

Memorizing classes will always be needed. If you take a pre-build framework or your own classes. The big productivity improvement is not in writing CSS using @apply and tailwind classes, but in writing just html and adding tailwind classes directly to the markup. So you need the backing css to already exist. Once you have to switch to a css file you're already loosing on the productivity gains. If you're using properly separated templates there might even be no need to later move the utilities to custom classes and `@apply` as everything's already in one place - the template of a component. Having the utilities out of the box is exactly the selling point of tailwind. Sass/less don't give you classes you can put in your markup - they're pre-processors. It also doesn't want to be a contender to e.g. uikit or bootstrap because they require so much work to be bent to custom design. Building your own components is expected, as custom design most often does require that anyways. There's however some work on a component library being done, which is build on top of tailwinds utility classes; and which will probably be easier to customize than e.g. bootstrap. If you have your own custom utilities library tailwind probably won't be of much benefit. If not however it' a great way to get started quickly on a existing, well documented, well adopted framework. Tailwind by default comes with 9 shades of each color, where afaik 4 are accessable colors on black and 5 are accessable on white background. And you can still edit them, those are just defaults. The message is not "you should not use more than (those) x colors/shades". The message is that a well managed design system must use a fixed set of colors, which is different to each person on the team just adding a new shade each and every time they add something to their codebase. The latter most often being the reason for the exploding number in color or text-size variations. It's about consistency and less about absolute numbers. Imho the need for all those customization variables of frameworks like uikit/bootstrap and the lock-in to a certain preprocessor already show how "inflexible" those solutions really are. I've not yet used any of those frameworks, where I didn't hit a brick wall at some point trying to skin a component to a custom design, because some aspect of a component wasn't customizable (either markup expectations or missing a "configuration variable"). To me tailwind css is neither an alternative to writing css, as the building blocks are way more cohesive and well rounded as well as "higher level" than css 1 – but also not for frameworks, which often impose quite an amount of markup constraints and understandably a certain kind of style, which can be problematic when implementing a custom design. [1] Take for example .truncate. There's no single css rule for truncating text properly. It's overflow: hidden; text-overflow: ellipsis; white-space: nowrap Same for e.g. .rounded-t. It applied rounded borders to both top edges, instead of css, which requires you to set it for each edge explicitly. This is more in line in how a designer would think about a design than how css needs it to be implemented. Nobody will say "make the top left and top right edge rounded". People say "make the top edges round" or "truncate that line".

Not a completly serious tip for listening during developing, but maybe a good one for listening before starting. ? or

I'm absolutely in awe at the speed you two are working at. I feel like a snail in comparison ? It's a separate thing so not one for right now but some similar code would be involved - the installer could do with reworking to allow easy presentation and installation of site profiles a bit like you can install a WP theme easily. You should be able to choose a site profile during installation (maybe thumbnails for each and click to popup the information about what it is and what it contains along with any caveats) and make that side of things easier too. I'm far from the only person to mention this, but I do know it's been mentioned for so many years now off and on and it lowers another barrier for entry in that people get to code that's relevant to them that little bit quicker. I think that one change, albeit not a simple one, would lead to more people producing site profiles as the current steps involved are again more numerous than they could to be and site profiles aren't paticularly visible as they are lumped with other modules in the directory. Also a note to say you guys probably want to make sure your tweaks to the modules interface don't show site profiles from the modules directory or that could get confusing.

i haven't looked at the functional fields yet... Here is my version of PGS: https://github.com/outflux3/ProcessGeneralSettings

With Version 2.5.0 they changed the class names of the HTML form to "typeahead__". I updated it in my original post, just so nobody gets confused.

I don't know how familiar you are with Process Modules, but I posted a quick example a while back. Once you get started, they are super addicting. You can build just about anything.