PROBLEMS

This file describes various problems that have been encountered
in compiling, installing and running GNU Emacs.

* Error compiling sysdep.c, "sioctl.h: no such file or directory".

Among USG systems with TIOCGWINSZ, some require sysdep.c to include
the file sioctl.h; on others, sioctl.h does not exist.  We don't know
how to distinguish these two kind of systems, so currently we try to
include sioctl.h on all of them.  If this #include gets an error, just
delete it.

* X Windows doesn't work if DISPLAY uses a hostname.

People have reported kernel bugs in certain systems that cause Emacs
not to work with X Windows if DISPLAY is set using a host name.  But
the problem does not occur if DISPLAY is set to `unix:0.0'.  I think
the bug has to do with SIGIO or FIONREAD.

You may be able to compensate for the bug by doing (set-input-mode nil nil).
However, that has the disadvantage of turning off interrupts, so that
you are unable to quit out of a Lisp program by typing C-g.

The easy way to do this is to put 

  (setq x-sigio-bug t)

in your site-init.el file.

* Watch out for .emacs files and EMACSLOADPATH environment vars

These control the actions of Emacs.
~/.emacs is your Emacs init file.
EMACSLOADPATH overrides which directories the function
"load" will search.

If you observe strange problems, check for these and get rid
of them, then try again.

* Fatal signal in the command  temacs -l loadup inc dump

This command is the final stage of building Emacs.  It is run by the
Makefile in the src subdirectory, or by build.com on VMS.

It has been known to get fatal errors due to insufficient swapping
space available on the machine.

On 68000's, it has also happened because of bugs in the
subroutine `alloca'.  Verify that `alloca' works right, even
for large blocks (many pages).

* test-distrib says that the distribution has been clobbered
* or, temacs prints "Command key out of range 0-127"
* or, temacs runs and dumps xemacs, but xemacs totally fails to work.
* or, temacs gets errors dumping xemacs

This can be because the .elc files have been garbled.  Do not be
fooled by the fact that most of a .elc file is text: these are
binary files and can contain all 256 byte values.

In particular `shar' cannot be used for transmitting GNU Emacs.
It typically truncates "lines".  What appear to be "lines" in
a binary file can of course be of any length.  Even once `shar'
itself is made to work correctly, `sh' discards null characters
when unpacking the shell archive.

I have also seen character \177 changed into \377.  I do not know
what transfer means caused this problem.  Various network
file transfer programs are suspected of clobbering the high bit.

The only verified ways to transfer GNU Emacs are `tar', kermit (in
binary mode on Unix), and rcp or internet ftp between two Unix systems,
or chaosnet cftp using raw mode.

If you have a copy of Emacs that has been damaged in its
nonprinting characters, you can fix them:

 1) Record the names of all the .elc files.
 2) Delete all the .elc files.
 3) Recompile alloc.c with a value of PURESIZE twice as large.
     You might as well save the old alloc.o.
 4) Remake xemacs.  It should work now.
 5) Running xemacs, do Meta-x byte-compile-file repeatedly
  to recreate all the .elc files that used to exist.
  You may need to increase the value of the variable
  max-lisp-eval-depth to succeed in running the compiler interpreted
  on certain .el files.  400 was sufficient as of last report.
 6) Reinstall the old alloc.o (undoing changes to alloc.c if any)
  and remake temacs.
 7) Remake xemacs.  It should work now, with valid .elc files.

* temacs prints "Pure Lisp storage exhausted"

This means that the Lisp code loaded from the .elc and .el
files during  temacs -l loadup inc dump  took up more
space than was allocated.

This could be caused by
 1) adding code to the preloaded Lisp files
 2) adding more preloaded files in loadup.el
 3) having a site-init.el or site-load.el which loads files.
   Note that ANY site-init.el or site-load.el is nonstandard;
   if you have received Emacs from some other site
   and it contains a site-init.el or site-load.el file, consider
   deleting that file.
 4) getting the wrong .el or .elc files
   (not from the directory you expected).
 5) deleting some .elc files that are supposed to exist.
   This would cause the source files (.el files) to be
   loaded instead.  They take up more room, so you lose.
 6) a bug in the Emacs distribution which underestimates
   the space required.

If the need for more space is legitimate, change the definition
of PURESIZE in config.h.

But in some of the cases listed above, this problem is a consequence
of something else that is wrong.  Be sure to check and fix the real
problem.

* Changes made to .el files do not take effect.

You may have forgotten to recompile them into .elc files.
Then the old .elc files will be loaded, and your changes
will not be seen.  To fix this, do M-x byte-recompile-directory
and specify the directory that contains the Lisp files.

* The dumped Emacs (xemacs) crashes when run, trying to write pure data.

Two causes have been seen for such problems.

1) On a system where getpagesize is not a system call, it is defined
as a macro.  If the definition (in both unexec.c and malloc.c) is wrong,
it can cause problems like this.  You might be able to find the correct
value in the man page for a.out (5).

2) Some systems allocate variables declared static among the
initialized variables.  Emacs makes all initialized variables in most
of its files pure after dumping, but the variables declared static and
not initialized are not supposed to be pure.  On these systems you
may need to add "#define static" to the m- or the s- file.

* Compilation errors on VMS.

You will get warnings when compiling on VMS because there are
variable names longer than 32 (or whatever it is) characters.
This is not an error.  Ignore it.

VAX C does not support #if defined(foo).  Uses of this construct
were removed, but some may have crept back in.  They must be rewritten.

There is a bug in the C compiler which fails to sign extend characters
in conditional expressions.  The bug is:
	char c = -1, d = 1;
	int i;

	i = d ? c : d;
The result is i == 255;  the fix is to typecast the char in the
conditional expression as an (int).  Known occurrences of such
constructs in Emacs have been fixed.

* rmail gets error getting new mail

rmail gets new mail from /usr/spool/mail/$USER using a program
called `movemail'.  This program interlocks with /bin/mail using
the protocol defined by /bin/mail.

There are two different protocols in general use.  One of them uses
the `flock' system call.  The other involves creating a lock file;
`movemail' must be able to write in /usr/spool/mail in order to do
this.  You control which one is used by defining, or not defining,
the macro MAIL_USE_FLOCK in config.h or the m- or s- file it includes.
IF YOU DON'T USE THE FORM OF INTERLOCKING THAT IS NORMAL ON YOUR
SYSTEM, YOU CAN LOSE MAIL!

If your system uses the lock file protocol, and fascist restrictions
prevent ordinary users from writing the lock files in /usr/spool/mail,
you may need to make `movemail' setgid to a suitable group such as
`mail'.

* Emacs won't work with X-windows if the value of DISPLAY is HOSTNAME:0.

Some people have found that Emacs was unable to connect to the local
host by name, as in DISPLAY=prep:0 if you are running on prep, but
could handle DISPLAY=unix:0.  Here is what tale@rpi.edu said:

      Seems as
    though gethostbyname was bombing somewhere along the way.  Well, we
    had just upgrade from SunOS 3.5 (which X11 was built under) to SunOS
    4.0.1.  Any new X applications which tried to be built with the pre
    OS-upgrade libraries had the same problems which Emacs was having.
    Missing /etc/resolv.conf for a little while (when one of the libraries
    was built?) also might have had a hand in it.

    The result of all of this (with some speculation) was that we rebuilt
    X and then rebuilt Emacs with the new libraries.  Works as it should
    now.  Hoorah.

* Emacs spontaneously displays "I-search: " at the bottom of the screen.

This means that Control-S/Control-Q "flow control" is being used.
C-s/C-q flow control is bad for Emacs editors because it takes away
C-s and C-q as user commands.  Since editors do not output long streams
of text without user commands, there is no need for a user-issuable
"stop output" command in an editor; therefore, a properly designed
flow control mechanism would transmit all possible input characters
without interference.  Designing such a mechanism is easy, for a person
with at least half a brain.

There are three possible reasons why flow control could be taking place:

  1) Terminal has not been told to disable flow control
  2) Insufficient padding for the terminal in use
  3) Some sort of terminal concentrator or line switch is responsible

First of all, many terminals have a set-up mode which controls
whether they generate flow control characters.  This must be
set to "no flow control" in order for Emacs to work.  Sometimes
there is an escape sequence that the computer can send to turn
flow control off and on.  If so, perhaps the termcap `ti' string
should turn flow control off, and the `te' string should turn it on.

Once the terminal has been told "no flow control", you may find it
needs more padding.  The amount of padding Emacs sends is controlled
by the termcap entry for the terminal in use, and by the output baud
rate as known by the kernel.  The shell command `stty' will print
your output baud rate; `stty' with suitable arguments will set it if
it is wrong.  Setting to a higher speed causes increased padding.  If
the results are wrong for the correct speed, there is probably a
problem in the termcap entry.  You must speak to a local Unix wizard
to fix this.  Perhaps you are just using the wrong terminal type.

For terminals that lack a "no flow control" mode, sometimes just
giving lots of padding will prevent actual generation of flow control
codes.  You might as well try it.

If you are really unlucky, your terminal is connected to the computer
through a concentrator which sends flow control to the computer, or it
insists on sending flow control itself no matter how much padding you
give it.  You are screwed!  You should replace the terminal or
concentrator with a properly designed one.  In the mean time,
some drastic measures can make Emacs semi-work.

One drastic measure to ignore C-s and C-q, while sending enough
padding that the terminal will not really lose any output.
Ignoring C-s and C-q can be done by using keyboard-translate-table
to map them into an undefined character such as C-^ or C-\.  Sending
lots of padding is done by changing the termcap entry.  Here is how
to make such a keyboard-translate-table:

    (let ((the-table (make-string 128 0)))
      ;; Default is to translate each character into itself.
      (let ((i 0))
	(while (< i 128)
	  (aset the-table i i)
	  (setq i (1+ i))))
      ;; Swap C-s with C-\
      (aset the-table ?\C-\\ ?\C-s)
      (aset the-table ?\C-s ?\C-\\)
      ;; Swap C-q with C-^
      (aset the-table ?\C-^ ?\C-q)
      (aset the-table ?\C-q ?\C-^)
      (setq keyboard-translate-table the-table))

An even more drastic measure is to make Emacs use flow control.
To do this, evaluate the Lisp expression (set-input-mode nil t).
Emacs will then interpret C-s and C-q as flow control commands.  (More
precisely, it will allow the kernel to do so as it usually does.)  You
will lose the ability to use them for Emacs commands.  Also, as a
consequence of using CBREAK mode, the terminal's Meta-key, if any,
will not work, and C-g will be liable to cause a loss of output which
will produce garbage on the screen.  (These problems apply to 4.2BSD;
they may not happen in 4.3 or VMS, and I don't know what would happen
in sysV.)  You can use keyboard-translate-table, as shown above,
to map two other input characters (such as C-^ and C-\) into C-s and
C-q, so that you can still search and quote.

I have no intention of ever redisigning the Emacs command set for
the assumption that terminals use C-s/C-q flow control.  This
flow control technique is a bad design, and terminals that need
it are bad merchandise and should not be purchased.  If you can
get some use out of GNU Emacs on inferior terminals, I am glad,
but I will not make Emacs worse for properly designed systems
for the sake of inferior systems.

* Control-S and Control-Q commands are ignored completely.

For some reason, your system is using brain-damaged C-s/C-q flow
control despite Emacs's attempts to turn it off.  Perhaps your
terminal is connected to the computer through a concentrator
that wants to use flow control.

You should first try to tell the concentrator not to use flow control.
If you succeed in this, try making the terminal work without
flow control, as described in the preceding section.

If that line of approach is not successful, map some other characters
into C-s and C-q using keyboard-translate-table.  The example above
shows how to do this with C-^ and C-\.

* Screen is updated wrong, but only on one kind of terminal.

This could mean that the termcap entry you are using for that
terminal is wrong, or it could mean that Emacs has a bug handing
the combination of features specified for that terminal.

The first step in tracking this down is to record what characters
Emacs is sending to the terminal.  Execute the Lisp expression
(open-termscript "./emacs-script") to make Emacs write all
terminal output into the file ~/emacs-script as well; then do
what makes the screen update wrong, and look at the file
and decode the characters using the manual for the terminal.
There are several possibilities:

1) The characters sent are correct, according to the terminal manual.

In this case, there is no obvious bug in Emacs, and most likely you
need more padding, or possibly the terminal manual is wrong.

2) The characters sent are incorrect, due to an obscure aspect
 of the terminal behavior not described in an obvious way
 by termcap.

This case is hard.  It will be necessary to think of a way for
Emacs to distinguish between terminals with this kind of behavior
and other terminals that behave subtly differently but are
classified the same by termcap; or else find an algorithm for
Emacs to use that avoids the difference.  Such changes must be
tested on many kinds of terminals.

3) The termcap entry is wrong.

See the file etc/TERMS for information on changes
that are known to be needed in commonly used termcap entries
for certain terminals.

4) The characters sent are incorrect, and clearly cannot be
 right for any terminal with the termcap entry you were using.

This is unambiguously an Emacs bug, and can probably be fixed
in termcap.c, tparam.c, term.c, scroll.c, cm.c or dispnew.c.

* Output from Control-V is slow.

On many bit-map terminals, scrolling operations are fairly slow.
Often the termcap entry for the type of terminal in use fails
to inform Emacs of this.  The two lines at the bottom of the screen
before a Control-V command are supposed to appear at the top after
the Control-V command.  If Emacs thinks scrolling the lines is fast,
it will scroll them to the top of the screen.

If scrolling is slow but Emacs thinks it is fast, the usual reason is
that the termcap entry for the terminal you are using does not
specify any padding time for the `al' and `dl' strings.  Emacs
concludes that these operations take only as much time as it takes to
send the commands at whatever line speed you are using.  You must
fix the termcap entry to specify, for the `al' and `dl', as much
time as the operations really take.

Currently Emacs thinks in terms of serial lines which send characters
at a fixed rate, so that any operation which takes time for the
terminal to execute must also be padded.  With bit-map terminals
operated across networks, often the network provides some sort of
flow control so that padding is never needed no matter how slow
an operation is.  You must still specify a padding time if you want
Emacs to realize that the operation takes a long time.  This will
cause padding characters to be sent unnecessarily, but they do
not really cost much.  They will be transmitted while the scrolling
is happening and then discarded quickly by the terminal.

Most bit-map terminals provide commands for inserting or deleting
multiple lines at once.  Define the `AL' and `DL' strings in the
termcap entry to say how to do these things, and you will have
fast output without wasted padding characters.  These strings should
each contain a single %-spec saying how to send the number of lines
to be scrolled.  These %-specs are like those in the termcap
`cm' string.

You should also define the `IC' and `DC' strings if your terminal
has a command to insert or delete multiple characters.  These
take the number of positions to insert or delete as an argument.

A `cs' string to set the scrolling region will reduce the amount
of motion you see on the screen when part of the screen is scrolled.

* You type Control-H (Backspace) expecting to delete characters.

Put `stty dec' in your .login file and your problems will disappear
after a day or two.

The choice of Backspace for erasure was based on confusion, caused by
the fact that backspacing causes erasure (later, when you type another
character) on most display terminals.  But it is a mistake.  Deletion
of text is not the same thing as backspacing followed by failure to
overprint.  I do not wish to propagate this confusion by conforming
to it.

For this reason, I believe `stty dec' is the right mode to use,
and I have designed Emacs to go with that.  If there were a thousand
other control characters, I would define Control-h to delete as well;
but there are not very many other control characters, and I think
that providing the most mnemonic possible Help character is more
important than adapting to people who don't use `stty dec'.

If you are obstinate about confusing buggy overprinting with deletion,
you can redefine Backspace in your .emacs file:
  (global-set-key "\b" 'delete-backward-char)
You may then wish to put the function  help-command  on some
other key.  I leave to you the task of deciding which key.

* Editing files through RFS gives spurious "file has changed" warnings.
It is possible that a change in Emacs 18.37 gets around this problem,
but in case not, here is a description of how to fix the RFS bug that
causes it.

    Date: Wed, 28 Jan 87 15:18:40 EST
    From: Bill Sommerfeld <wesommer@athena.mit.edu>
    Sender: wesommer@athena.mit.edu
    To: toddb%crl.tek.csnet@relay.cs.net
    Cc: bug-rfs, bug-gnu-emacs@prep.ai.mit.edu
    Subject: Bug in RFS server (fsync())

    There was a serious pair of bugs in the handling of the fsync() system
    call in the RFS server.

    The first is that the fsync() call is handled as another name for the
    close() system call (!!).  It appears that fsync() is not used by very
    many programs; Emacs version 18 does an fsync() before closing files
    to make sure that the bits are on the disk.

    This is fixed by the enclosed patch to the RFS server.

    The second, more serious problem, is that fsync() is treated as a
    non-blocking system call (i.e., it's implemented as a message that
    gets sent to the remote system without waiting for a reply).  Fsync is
    a useful tool for building atomic file transactions.  Implementing it
    as a non-blocking RPC call (when the local call blocks until the sync
    is done) is a bad idea; unfortunately, changing it will break the RFS
    protocol.  Any suggestions?

					    Bill Sommerfeld
					    MIT Project Athena

    (as always, your line numbers may vary)

    % rcsdiff -c -r1.2 serversyscall.c
    RCS file: RCS/serversyscall.c,v
    retrieving revision 1.2
    diff -c -r1.2 serversyscall.c
    *** /tmp/,RCSt1003677   Wed Jan 28 15:15:02 1987
    --- serversyscall.c     Wed Jan 28 15:14:48 1987
    ***************
    *** 163,169 ****
	    /*
	     * No return sent for close or fsync!
	     */
    !       if (syscall == RSYS_close || syscall == RSYS_fsync)
		    proc->p_returnval = deallocate_fd(proc, msg->m_args[0]);
	    else
	    {
    --- 166,172 ----
	    /*
	     * No return sent for close or fsync!
	     */
    !       if (syscall == RSYS_close)
		    proc->p_returnval = deallocate_fd(proc, msg->m_args[0]);
	    else
	    {

* ld complains because `alloca' is not defined on your system.

Alloca is a library function in 4.2bsd, which is used very heavily by
GNU Emacs.  Use of malloc instead is very difficult, as you would have
to arrange for the storage to be freed, and do so even in the case of
a longjmp happening inside a subroutine.  Many subroutines in Emacs
can do longjmp.

If your system does not support alloca, try defining the symbol
C_ALLOCA in the m-...h file for that machine.  This will enable the use
in Emacs of a portable simulation for alloca.  But you will find that
Emacs's performance and memory use improve if you write a true
alloca in assembler language.

alloca (N) should return the address of an N-byte block of memory
added dynamically to the current stack frame.

* Vax C compiler bugs affecting Emacs.

You may get one of these problems compiling Emacs:

   foo.c line nnn: compiler error: no table entry for op STASG
   foo.c: fatal error in /lib/ccom

These are due to bugs in the C compiler; the code is valid C.
Unfortunately, the bugs are unpredictable: the same construct
may compile properly or trigger one of these bugs, depending
on what else is in the source file being compiled.  Even changes
in header files that should not affect the file being compiled
can affect whether the bug happens.  In addition, sometimes files
that compile correctly on one machine get this bug on another machine.

As a result, it is hard for me to make sure this bug will not affect
you.  I have attempted to find and alter these constructs, but more
can always appear.  However, I can tell you how to deal with it if it
should happen.  The bug comes from having an indexed reference to an
array of Lisp_Objects, as an argument in a function call:
  Lisp_Object *args;
  ...
   ... foo (5, args[i], ...)...
putting the argument into a temporary variable first, as in
  Lisp_Object *args;
  Lisp_Object tem;
  ...
   tem = args[i];
   ... foo (r, tem, ...)...
causes the problem to go away.
The `contents' field of a Lisp vector is an array of Lisp_Objects,
so you may see the problem happening with indexed references to that.

* 68000 C compiler problems

Various 68000 compilers have different problems.
These are some that have been observed.

** Using value of assignment expression on union type loses.
This means that  x = y = z;  or  foo (x = z);  does not work
if x is of type Lisp_Object.

** "cannot reclaim" error.

This means that an expression is too complicated.  You get the correct
line number in the error message.  The code must be rewritten with
simpler expressions.

** XCONS, XSTRING, etc macros produce incorrect code.

If temacs fails to run at all, this may be the cause.
Compile this test program and look at the assembler code:

struct foo { char x; unsigned int y : 24; };

lose (arg)
     struct foo arg;
{
  test ((int *) arg.y);
}

If the code is incorrect, your compiler has this problem.
In the XCONS, etc., macros in lisp.h you must replace (a).u.val with
((a).u.val + coercedummy) where coercedummy is declared as int.

This problem will not happen if the m-...h file for your type
of machine defines NO_UNION_TYPE.  That is the recommended setting now.

* C compilers lose on returning unions

I hear that some C compilers cannot handle returning
a union type.  Most of the functions in GNU Emacs return
type Lisp_Object, which is currently defined as a union.

This problem will not happen if the m-...h file for your type
of machine defines NO_UNION_TYPE.  That is the recommended setting now.