should this be flush() instead of end, because there is no rule not adding antoher thing after reading.

this looks very implictit and hard to understand what is going on, maybe a function to HtmlWriter to handle this directly, because we'll use this all over the place, but the disadvante is that we add Grantlee dependes to mimetreeparser, so it's a no-go

htmlWriter->addGrantleeOutput(t, &c);

so maybe better add a util method:

AddGrantlleOutputToHtmlWriter(t, &c, htmlWriter);

shouldn't we still flush, at the end?

you removed everywhere the flush but add in the docu, that it needs to be flushed, that sounds wired

Theoretically, yes. However that makes no sense semantically, as you would duplicate the first part of the output.

Just to save one line of code? And I don't see how that makes anything clearer, "OutputStream" looks pretty self-explanatory to me.

HtmlWriter::end() flushes the stream, closing the file flushes the device, so that's all covered.

That's referring to QTextStream::flush() that you can call on stream() directly. For normal QString output you wont need that, just for raw device access.

well for me it is much more obvious what is going on and i think a extra function would help understanding the code faster. and i think the compiler will mark this function as inline, so no overhead.
I also have in mind, that we may need to adjust this in future, or anything like that.
What about, that the whole block with isAttachment put in one function?

Factoring out the whole attachment block seems more useful indeed. But that's IMHO out of scope for this commit. Let's revisit cleaning up the render implementation once we have the API sorted out.