bug#48100: 28.0.50; inserting too many lines into a fresh cpp file breaks the buffer

[Alan Mackenzie]

Hello, Paul and Lars.

On Thu, Apr 29, 2021 at 10:04:36 +0200, Paul Nelson wrote:
> Inserting a sufficiently large chunk of text into a fresh cpp file
> breaks syntax highlighting and causes the buffer to enter a
> semi-permanent broken state.  Moreover, the offending function
> "c-guess-basic-syntax" seems to break edebug instrumentation.

Thanks for taking the trouble to report these bugs.

> In more detail:

> Start from emacs -Q.  Insert sufficiently many lines of C++ code in a
> temporary buffer.  A slightly contrived example:

> #include <cstdio>
> #include <cstdlib>
> #include <cstring>
> #include <climits>
> #include <cfloat>
> #include <vector>
> #include <cmath>
> #include <set>
> #include <list>
> #include <map>
> #include <stack>
> #include <queue>
> #include <iostream>
> #include <sstream>
> #include <fstream>
> #include <algorithm>
> #include <numeric>
> #include <functional>
> #include <cstdio>
> #include <cstdlib>
> #include <cstring>
> #include <climits>
> #include <cfloat>
> #include <vector>
> #include <cmath>
> #include <set>
> #include <list>
> #include <map>
> #include <stack>

> Copy these lines to the kill ring.  Create a new file "test.cpp", and
> paste what was copied.

I'm assuming here that by "paste" you mean type C-y.

> The first error is that syntax highlighting stops after a certain
> number of lines (in the above example, just before the final line).

I can't reproduce this, that's on GNU/Linux, neither in a tty nor in
X-Windows.

However, it looks like on your machine Emacs is fontifying one single
500-byte chunk (which is rounded up to the next EOL) and then neglecting
the rest of the lines input (i.e. the final line with <stack>).

> The same problem arises if such lines are inserted via abbrev-mode.
> This is the intended use case.

> No problem if such lines are copy/pasted in small enough chunks.

> The problem *appears* to go away if one cuts and then pastes the
> offending lines, but the buffer remains in a broken state.  For
> instance, attempting to insert a couple newlines followed by
>   int main(
> gives the following error:

> Debugger entered--Lisp error: (wrong-type-argument number-or-marker-p nil)
>   c-guess-basic-syntax()
>   c-indent-line()
>   indent-according-to-mode()
>   c-electric-paren(nil)
>   funcall-interactively(c-electric-paren nil)
>   command-execute(c-electric-paren)

I can't reproduce that either, I'm afraid.  At least, not yet.

> Incidentally, "C-u C-M-x" applied to c-guess-basic-syntax gives a long and
> complicated backtrace that starts with:

> Debugger entered--Lisp error: (invalid-read-syntax "Expected" "lambda")
>   edebug-syntax-error("Expected" "lambda")

That, I can reproduce.  After a bit of experimentation, it would appear
that this is due to the lack of an edebug-spec for the macro
c-save-buffer-state.  To confirm this in the most basic manner, type

   M-: (def-edebug-spec c-save-buffer-state let*)

, after which C-u C-M-x works on c-guess-basic-syntax.  More
practically, this situation can be worked around with

  M-x load-library RET cc-mode RET

, which loads cc-defs.elc, which contains the pertinent edebug-spec.  I
don't understand how that edebug-spec in cc-defs.elc is not loaded
anyway, given the you're already using CC Mode.  Perhaps that will
become clearer in the next few days.

> In GNU Emacs 28.0.50 (build 2, x86_64-apple-darwin19.6.0, NS appkit-1894.60
> Version 10.15.7 (Build 19H524))
>  of 2021-04-26 built on Pauls-MBP-2
> Repository revision: 9d34fd8b33c55768190d41239931120e3fc9717f
> Repository branch: master
> Windowing system distributor 'Apple', version 10.3.1894
> System Description:  Mac OS X 10.15.7

> Configured using:
>  'configure --with-native-compilation'

[ .... ]

-- 
Alan Mackenzie (Nuremberg, Germany).