PHP output buffering: When/whether to use for different kinds of real existing sites and applications?

Question

Working mainly in WordPress contexts, I have mostly done without output buffering in my PHP code, but I have recently begun to experiment with it and try to discern simply as a user whether there was any noticeable performance impact in typical (for me) scenarios. Can't say I've achieved anything definitive there, but I don't know that my experiments really apply to scaled-up situations.

My general impression is that, unless I have a very good reason for using it - probably involving the kind of script in which I'm not generally interested at this time - I should avoid it. Yet I'll read suggestions that I could use buffering in one or another normal context (e.g., a shortcode that renders a large block of HTML), and I often see it being used in code by people whose work I admire and try to emulate. I've seen the familiar ob_start() etc. sequences applied for rendering even very minimal output that is not further adjusted: a menu or other short list, for example.

I have also seen people assert that there is simply no need for output buffering in well-written code, except in peculiar situations. See StackOverflow Q&A's here Why use output buffering in PHP?, to give one example. Others will say it's useful ONLY when some manipulation of the output is necessary - like a preg_replace of content - but you don't need an output buffer to do that kind of thing (any ol' variable will do).

In most real world instances of the sort I encounter, some version of page caching will also be in use, and sometimes several different plug-ins each with its own caching. Considering how many functions will themselves employ numerous nested/secondary HTML-producing functions, you could easily produce thousands of little nested buffered blocks per page: Suppose a foreach loop produces n number of blog-comments within a larger page: Each blog-comment is rendered by a function, and could be buffered separately. The each-comment function could be called within the foreach loop of the function that outputs the comment thread, and its output could also be buffered. Then the function that outputs the post that includes the comment thread could be output-buffered before being rendered. The entire page, including sundry independently output-buffered templates, plug-ins, widgets, menus, etc., etc., could also be output-buffered.

At what point would any of it be useful or, alternatively, completely counterproductive?

If the answer involves any element of "well, you wouldn't notice in your blog, but you'd need it for BuzzFeed or the New York Times," or the reverse, how would I estimate when the point has been reached?

At this point, I'm leaning toward writing the code as cleanly as possible, and letting (what I understand to be) built-in PHP, browser, server, etc., buffering and explicit caching functions do the work of making the page or its elements load efficiently and quickly, which is the real objective, no?

Answer 1

When using template engines, the code is typically output just once — after the whole page code has been generated. So output buffering is generally unneeded for such case.

The only situation when output buffering is really inevitable is capturing output of functions like var_dump() or phpinfo() without sending it to browser. Also, this can be useful for some third-party libraries that use echo without providing ability to get the result directly as a string.

Answer 2

I'll post this as an answer to my own question, though not under the assumption that there are no better ones, as I have run across a circumstance that might even qualify as a common for a certain type of plug-in, including a type that I am, in fact, quite interested in. The answer applies mainly to WordPress and to making plug-ins extensible.

In brief, in order to provide a filter - using the common WordPress "apply_filters" function - you will frequently need to be able to capture the entirety of your HTML output at once. When creating a large, complex block with mixed HTML and secondary functions, using an ob_start()/ob_get_[] sequence will be markedly more efficient way of achieving this end.

So: A plugin on which I'm currently working outputs a table whose elements are put together using a number of secondary functions. Now, I could conceivable do the entire thing adding content to a single variable, but it's much easier and more economical to initiate output buffering at the beginning, and then write the code without the extra layer of abstraction of $html .= [more and more HTML and PHP] In the end, the code assigns the entire output to a variable - $html = ob_get_clean(); and then returns the variable with the apply_filters function: return apply_filters( 'filter-tag' , $html ); Additionally, ob_start() and ob_get_clean() mark points in the code where it makes sense to provide do_action hooks.