Hi Nicolas,
    I've included the simplest patch I could come up with for the
divergence in behavior between org-babel-tangle-single-file and
org-link-search. I think there are two new threads that I need to
create. One is related to how to make it possible to specify what
should be removed along with the coderef (i.e. coderef prefix), the
other is the addition of header arguments that provide the same
functionality as switches. Best,
Tom

> This is already conflating the two. I'd like to solve the issue at hand
> without having header args interfere at all.
>
> This can happen later, after a discussion on the ML.

Ok. I've included the simplest version of the fix, which is to use
org-src-coderef-regexp in org-babel-tangle-single-file.

> Would you mind answering my questions first? I still don't follow you
> about the coderef prefix/regexp.

https://code.orgmode.org/bzg/org-mode/src/2d78ea57cfad1ddc3e993c949daf117b76315170/lisp/org-src.el#L882

That line defines a hardcoded regular expression for matching
coderefs. The codref prefix is the first =[ \t]*= and the coderef
regexp is the equivalent to the fully formatted version of that format
string. Neither of those can currently be specified by the user. The
user should not be able to specify the coderef regexp due to the fact
that it is too easy to specify a regexp that will not work correctly
and because the format string is needed to make org-link-search work
for named coderefs (otherwise you wind up trying to replace .+ in the
coderef regexp which is a nightmare). The coderef prefix is something
that should probably be configurable by the user so that empty
comments are not left in the file. I also looked into detecting the
comment character for the language in question, but that is
significantly more difficult even using (with-temp-buffer (funcall
lang-mode) comment-start) because not all languages have sane comment
start values and comment-start is not complete, so we would need a way
to manually specify what to exclude anyway.