SQA:
Some of this is a little pedantic these days. The need for accessing old
versions of UNIX is less now that Linux and BSD exist on PCs.
Unsigned char:
Use 'unsigned char', never 'char'. This is a big pain, but it
prevents char to int when you meant unsigned char to int conversion
bugs.
gcc:
Please run gcc -Wall. Make especially sure that there are no
functions used before defined warnings. These are OK on 32-bit
systems, but break on 64-bit systems where sizeof(int) doesn't equal
sizeof(void *).
Buffer and file offsets:
Use 'long'. We should really be using 'unsigned long'.
Valgrind:
Please do some checking with 'valgrind'. This tool finds accesses
to uninitialized memory. It also finds memory leaks.
debug_joe:
Please hit ESC x debug_joe a few times. Make sure there are no P
leaks (forgot to call prm()).
It would be nice:
If vs type was different from z-string type.
Dangerous situations:
assume maint->curwin->object is a BW *
call interactive functions (like doedit) and expect them
to leave maint a buffer window (it could start a prompt).
should check plain file checking.
vs, zstring, cstring, there are too many, each with its own
memory management.
These should be relaxed. Nobody uses ancient C compilers any more:
Include files:
You can rely on:
stdio.h, string.h, errno.h and math.h
Everything else needs #ifdefs and a check in the configure script.
Really old systems did not have stdlib.h or unistd.h. Non UNIX
systems don't necessarily have fcntl.h or sys/stat.h.
#if:
Use #ifdef, not #if. Old systems did not have #if. For example,
use #ifdef junk instead of #if 0.
PARAMS():
This should help porting JOE to really old systems (except that the
configure script probably wont work). However, declarations allow
the compiler to automatically insert casts. This can't happen if
they are not there. This has different consequences, depending
on the word size:
32-bit systems:
functions should not define float args, only double.
should not rely on prototypes for double to long conversions.
Systems where int is not the same as long (16-bit systems) will
have lots of trouble without declarations.
------------
Edit Buffers
------------
API:
Look at the comments in b.h for more information.
B *bfind(unsigned char *name);
Load disk file into memory buffer 'B'.
bsave(P *p,unsigned char *name,int size);
Write size bytes from buffer beginning at p to disk file
brm(b); Free data structure
Once you have a B you can access the characters in it via P pointers (which
are like C++ STL iterators).
B *b = bfind("foo"); Load file into memory
P *p = pdup(b->bof); Get pointer to beginning of file (duplicate
b->bof which is a P).
prm(p); Free p when we're done with it.
int c=brch(p); Get character at p.
int c=pgetc(p); Get character at p and advance it.
int c=prgetc(p); Retract p, then return character at it.
- These return -1 (NO_MORE_DATA) if you try to read end of file or
before beginning of file.
- A pointer can point to any character of the file and right after the
end of the file.
- For UTF-8 files, character can be between 0 and 0x7FFFFFFF
Publicly readable members of P:
p->byte The byte offset into the buffer
p->line The line number
p->xcol If P is the cursor, this is the column number
where the cursor will be displayed on the screen
(which does not have to match the column of the
character at P).
Some predicates:
pisbof(p); True if pointer is at beginning of buffer
piseof(p); True if pointer is at end of buffer
pisbol(p); True if pointer is at beginning of line
piseol(p); True if pointer is at end of line
pisbow(p); True if pointer is at beginning of a word
piseow(p); True if pointer is at end of a word
More information about character at p:
piscol(p); Get column number of character at p.
Some other ways of moving a P through a B:
pnextl(p); Go to beginning of next line
pprevl(p); Go to end of previous line
pfwrd(p,int n); Move forward n characters
pbkwd(p,int n); Move backward n characters
p_goto_bof(p);
p_goto_eof(p);
p_goto_bol(p);
p_goto_eol(p);
pset(p,q); Move p to same position as q.
pline(p,n); Goto to beginning of a specific line.
pgoto(p,n); Goto a specific byte position.
pfind(P,unsigned char *s,int len);
Fast Boyer-Moore search forward.
prfind(P,unsigned char *s,int len);
Fast Boyer-Moore search backward.
These are very fast- they look at low level
data structure and don't go through pgetc(). Boyer-Moore
allows you to skip over many characters without reading
them, so you can get something like O(n/len).
Some facts:
Local operations are fast: pgetc(), prgetc().
Copy is fast: pset().
pline() and pgoto() are slower, but look for the closest existing
P to start from.
The column number is stored in P, but it is only updated if
it is easy to do so. If it's hard (like you crossed a line
boundary backward) it's marked as invalid. piscol() then has
to recalculate it.
Modifying a buffer:
binsc(p,int c); Insert character at p.
bdel(P *from,P *to); Delete character between two Ps.
Note that when you insert or delete, all of the Ps after the insertion/
deletion point are adjusted so that they continue to point to the same
characeter before the insert or delete.
Insert and Delete create undo records.
Insert and Delete set dirty flags on lines which are currently being
displayed on the screen, so that when you return to the edit loop, these
lines automatically get redrawn.
Internal:
An edit buffer is made up of a doubly-linked list of fixed sized (4 KB)
gap buffers. A gap buffer has two parts: a ~16 byte header, which is always
in memory, and the actual buffer, which can be paged out to a swap file (a
vfile- see vfile.h). A gap buffer consists of three regions: text before
the gap, the gap and text after the gap (which always goes all the way to
the end of buffer). (hole and ehole in the header indicate the gap
position). The size of the gap may be 0 (which is the case when a file is
first loaded). Gap buffers are fast for small inserts and deletes when the
cursor is at the gap (for delete you just adjust a pointer, for insert you
copy the data into gap). When you reposition the cursor, you have to move
the gap before any inserts or deletes occur. If you had only one window and
a single large gap buffer for the file, you could always keep the gap at the
cursor- the right-arrow key copies one character across the gap.
Of course for edits which span gap buffers or which are larger than a gap
buffer, you get a big mess of gap buffer splitting and merging plus
doubly-linked list splicing.
Still, this buffer method is quite fast: you never have to do large memory
moves since the gap buffers are limited in size. To help search for line
numbers, the number of newlines '\n's contained in each gap buffer is stored
in the header. Reads are fast as long as you have a P at the place you
want to read from, which is almost always the case.
It should be possible to quickly load files by mapping them directly into
memory (using mmap()) and treating each 4KB page as a gap buffer with 0 size
gap. When page is modified, use copy-on-write to move the page into the
swap file (change pointer in header). This is not done now. Instead the
file is copied when loaded.
----------------
Windowing System
----------------
There is a tiny object-oriented windowing system built into JOE. This is
the class hierarchy:
SCRN
A optimizing terminal screen driver (very similar to 'curses').
has a pointer to a CAP, which has the terminal capabilities read
from termcap or terminfo.
writes output to screen with calls to the macro ttputc(). (tty.c is the
actual interface to the tty device).
cpos() - set cursor position
outatr() - draw a character at a screen position with attributes
eraeol() - erase from some position to the end of the line
SCREEN
Contains list of windows on the screen (W *topwin).
Points to window with focus (W *curwin).
Contains pointer to a 'SCRN', the tty driver for the particular terminal
type.
W
A window on a screen.
Has position and size of window.
Has:
void *object- pointer to a structure which inherits window (W should
really be a base class for these objects- since C doesn't have this
concept, a pointer to the derived class is used instead- the derived
class has a pointer to the base class: it's called 'parent').
Currently this is one of:
BW * a text buffer window (screen update code is here.)
QW * query window (single character yes/no questions)
MENU * file selection menu
BW * is inherited by (in the same way that a BW inherits a W):
PW * a single line prompt window (file name prompts)
TW * a text buffer window (main editing window).
WATOM *watom- Gives type of this window. The WATOM structure has
pointers to virtual member functions.
KBD *kbd- The keyboard handler for this window. When window has
focus, keyboard events are sent to this object. When key sequences
are recognized, macros bound to them are invoked.
Some window are operators on others. For example ^K E, load a file into a
window prompt operates on a text window. If you hit tab, a file selection
menu which operates on the prompt window appears below this. When a window
is the target of operator windows is killed, the operators are killed also.
Currently all windows are currently the width of the screen (should be fixed
in the future). The windows are in a big circular list (think of a big loop
of paper). The screen is small window onto this list. So unlike emacs, you
can have windows which exist, but which are not on the screen.
^K N and ^K P move the cursor to next or previous window. If the next
window is off the screen it is moved onto the screen, along with any
operator windows are target it.
------
MACROS
------
- add something here.
-------------
Screen update
-------------
- add something here.
-----
Files
-----
main.c has main().
b.c Text buffer management
undo.c Undo system.
kbd.c Keymap datastructure (keysequence to macro bindings).
macro.c Keyboard and joerc file macros
help.c Implement the on-line help window
poshist.c Cursor position history
rc.c joerc file loader
tab.c tab completion for file selection prompt
regex.c regular expressions
blocks.c Library: fast memcpy() functions (necessary on really old versions of UNIX).
dir.c Directory reading functions (for old UNIXs).
hash.c Library: simple hash functions.
vs.c Automatic variable length strings (like C++ string).
va.c Automatic array of strings (like STL container)
vfile.c Library: virtual memory functions (allows you to edit files larger than memory)
utils.c Misc. utilities
queue.c Library: doubly linked lists
path.c Library: file name and path manipulation functions
selinux.c secure linux functions
i18n.c Unicode character type information database
charmap.c UNICODE to 8-bit conversion functions
utf8.c UTF-8 to unicode coding functions
termcap.c load terminal capabilities from /etc/termcap file or terminfo database
scrn.c terminal update functions (curses)
syntax.c syntax highlighter
cmd.c Table of user edit functions
ublock.c User edit functions: block moves
uedit.c User edit functions: basic edit functions
uerror.c User edit functions: parse compiler error messages and goto next error, previous error
ufile.c User edit functions: load and save file
uformat.c User edit functions: paragraph formatting, centering
uisrch.c User edit functions: incremental search
umath.c User edit functions: calculator
usearch.c User edit functions: search & replace
ushell.c User edit functions: subshell
utag.c User edit functions: tags file search
menu.c A class: menu windows
tw.c A class: main text editing window
qw.c A class: query windows
pw.c A class: prompt windows
bw.c A class: text buffer window (screen update code is here)
w.c A class: base class for all windows
-------
Strings
-------
char * C-strings: only used for system calls or C-library calls.
unsigned char * Z-strings: used in JOE for read-only code
vs s V-strings: exist in heap
s.c_string Get C-string out of it (0 time)
z.z_string Get Z-string out of it (0 time)
vsrm(&s); Free a vs.
vs n=vsdup(s) Duplicate a vs.
vsadd(&s, 'c') Append one character.
vscat(&s, zs, int len) Concatenate array on end of string
vscat(&s, sc("Hi there"))
vscat(&s, sv(s))
vscat(&s, sz(z))
vscmp()
