The implementation does not seem to do what you describe. It still uses the HtmlDumper for all non-CLI SAPIs but uses it also for some cases of CLI SAPIs.

I'm not sure to get what you mean. The goal is to use CliDumper if we're in a CLI env, just as before. But if we're in another env or the Accept header is present with text/html, then we use HtmlDumper.

If you debug a JSON API, you don't want to HTML output because it messes with the output.

This is not what your new condition does

I'm sorry, I don't get where the condition is wrong 😕 Could you maybe precisely point at the mistake please?

It still uses the HtmlDumper for all non-CLI SAPIs

It is the current behavior. All this PR does is "forcing" the HTML output in case of the Accept header.

@alexandre-daubois in your condition, anything that is not a CLI SAPI will still use the HtmlDumper. what you changed is that you use the HtmlDumper in more cases, not less cases.

Alright, I think I figured out the problem, thank you for your explanation. Is it better now? If the header is present with the good value, we chose HtmlDumper. Otherwise, we fallback on CliDumper (just like proposed in #54772 (comment) actually).

At this point, I'm not 100% sure if this should be considered a bug or a feature 🤔

I would treat it as a feature.

It is now updated with your suggestions, thank you!

I would recommend to create a new "plain text only" dumper, and use that as the default fallback (if no accept html header is present)
HtmlDumper and CliDumper are imho not good fallbacks for most usecases

I would recommend to create a new "plain text only" dumper, and use that as the default fallback (if no accept html header is present) HtmlDumper and CliDumper are imho not good fallbacks for most usecases

You are right, the CliDumper does not output anything if it is not in a terminal. I tried to see what would be the result of CliDumper in Chrome dev tool and it is just empty.

Thus the problem I reported in 54772 (and probably the one described in 49754 as well) is not solved by using CLiDumper instead of HtmlDumper. A plain text dumper is indeed necessary.

@pierresh I think that your case could (and should) be solved by launching a local VarDumper server so the output doesn't messes up with your API response

@pierresh I think that your case could (and should) be solved by launching a local VarDumper server so the output doesn't messes up with your API response

a dd() will (and should) always mess up the output, no one excepts valid formatted json output if they call dd(), else you would return a new JsonResponse() instead of calling dd().

Also when debugging production / staging / docker etc, its not always possible to launch a VarDumper server.

I'm not convinced that using dd() is the most suitable solution for debugging a production application. Apart from that, have you looked at the static property AbstractDumper::$defaultOutput, but also AbstractDumper::setOutput()? You can pass any callable/resource/output-path to it, which may solve your issue.

I think that the conversation is moving away from the original purpose of the PR. Perhaps it's a good idea to continue the discussion in the respective issues to avoid any confusion? 🙂

Does this really work on Symfony apps? Don't we register another handler that'd also need a similar logic, but that'd use the request_stack instead of the globals to get the accept header?

I tried with a new Symfony app and a dummy controller like the following:

class HomeController
{
    #[Route(path: '/')]
    public function home(): Response
    {
        dd(PHP_SAPI);
    }
}

It dumped the value in the CLI, and dumped as HTML in the browser when adding Accept: text/html. Is there a place or a case I'm missing?