lua/bugs

--[[
** lua.stx / llex.c
Tue Dec  2 10:45:48 EDT 1997
>> BUG: "lastline" was not reset on function entry, so debug information
>> started only in the 2nd line of a function.



=================================================================
--- Version 3.1 alpha

** lua.c
Thu Jan 15 14:34:58 EDT 1998
>> must include "stdlib.h" (for "exit()").

** lbuiltin.c / lobject.h
Thu Jan 15 14:34:58 EDT 1998
>> MAX_WORD may be bigger than MAX_INT
(by lhf)

** llex.c
Mon Jan 19 18:17:18 EDT 1998
>> wrong line number (+1) in error report when file starts with "#..."

** lstrlib.c
Tue Jan 27 15:27:49 EDT 1998
>> formats like "%020d" were considered too big (3 digits); moreover,
>> some sistems limit printf to at most 500 chars, so we can limit sizes
>> to 2 digits (99).

** lapi.c
Tue Jan 27 17:12:36 EDT 1998
>> "lua_getstring" may create a new string, so should check GC

** lstring.c / ltable.c
Wed Jan 28 14:48:12 EDT 1998
>> tables can become full of "empty" slots, and keep growing without limits.

** lstrlib.c
Mon Mar  9 15:26:09 EST 1998
>> gsub('a', '(b?)%1*' ...) loops (because the capture is empty).

** lstrlib.c
Mon May 18 19:20:00 EST 1998
>> arguments for "format" 'x', 'X', 'o' and 'u' must be unsigned int.



=================================================================
--- Version 3.1

** liolib.c / lauxlib.c
Mon Sep  7 15:57:02 EST 1998
>> function "luaL_argerror" prints wrong argument number (from a user's point
of view) when functions have upvalues.

** lstrlib.c
Tue Nov 10 17:29:36 EDT 1998
>> gsub/strfind do not check whether captures are properly finished.
(by roberto/tomas)

** lbuiltin.c
Fri Dec 18 11:22:55 EDT 1998
>> "tonumber" goes crazy with negative numbers in other bases (not 10),
because "strtol" returns long, not unsigned long.
(by Visual C++)

** lstrlib.c
Mon Jan  4 10:41:40 EDT 1999
>> "format" does not check size of format item (such as "%00000...00000d").

** lapi.c
Wed Feb  3 14:40:21 EDT 1999
>> getlocal cannot return the local itself, since lua_isstring and
lua_isnumber can modify it.

** lstrlib.c
Thu Feb  4 17:08:50 EDT 1999
>> format "%s" may break limit of "sprintf" on some machines.
(by Marcelo Sales)

** lzio.c
Thu Mar  4 11:49:37 EST 1999
>> file stream cannot call fread after EOF.
(by lhf)



=================================================================
--- Version 3.2 (beta)

** lstrlib.c
Fri Apr 30 11:10:20 EST 1999
>> '$' at end of pattern was matching regular '$', too.
(by anna; since 2.5)

** lbuiltin.c
Fri May 21 17:15:11 EST 1999
>> foreach, foreachi, foreachvar points to function in stack when stack
can be reallocated.
(by tomas; since 3.2 beta)

** lparser.c
Wed Jun 16 10:32:46 EST 1999
>> cannot assign to unlimited variables, because it causes overflow in
the number of returns of a function.
(since 3.1)



=================================================================
--- Version 3.2

** lmathlib.c
Wed Aug 18 11:28:38 EST 1999
>> random(0) and random(x,0) are wrong (0 is read as no argument!).
(by Dave Bollinger; since 3.1)

** lparser.c
Thu Sep  2 10:07:20 EST 1999
>> in the (old) expression << ls->fs->f->consts[checkname(ls)] >>, checkname
could realloc f->consts.
(by Supratik Champati; since 3.2 beta)

** lobject.c / lbuiltin.c
Wed Sep  8 17:41:54 EST 1999
>> tonumber'e1' and tonumber('  ', x), for x!=10, gave 0 instead of nil.
(since 3.1)

** lstrlib.c
Thu Nov 11 14:36:30 EDT 1999
>> `strfind' does not handle \0 in plain search.
(by Jon Kleiser; since 3.1)

** lparser.c
Wed Dec 29 16:05:43 EDT 1999
>> return gives wrong line in debug information
(by lhf; since 3.2 [at least])

** ldo.c
Thu Dec 30 16:39:33 EDT 1999
>> cannot reopen stdin (for binary mode)
(by lhf & roberto; since 3.1)

** lapi.c
Thu Mar  2 09:41:53 EST 2000
>> lua_settable should check stack space (it could call a T.M.)
(by lhf & celes; since 3.2; it was already fixed by fixed stack)

** lparser.c
Mon Apr  3 09:59:06 EST 2000
>> '%' should be in expfollow
(by Edgar Toernig; since 3.1; it was already fixed)

** lbuiltin.c
Mon Apr  3 10:05:05 EST 2000
>> tostring() without arguments gives seg. fault.
(by Edgar Toernig; since 3.0)



=================================================================
--- Version 4.0 alpha

Tested with full test suites (as locked in Mon Apr 24 14:23:11 EST 2000)
in the following platforms:
* Linux   - gcc, g++
* AIX     - gcc
* Solaris - gcc, cc
* IRIX	  - cc, cc-purify
* Windows - Visual C++ (.c e .cpp, warning level=4)


** lstrlib.c
Tue May  2 15:27:58 EST 2000
>> `strfind' gets wrong subject length when there is an offset
(by Jon Kleiser; since 4.0a)

** lparser.c
Fri May 12 15:11:12 EST 2000
>> first element in a list constructor is not adjusted to one value
>> (e.g. <20>a = {gsub('a','a','')}<7D>)
(by Tomas; since 4.0a)

** lparser.c
Wed May 24 14:50:16 EST 2000
>> record-constructor starting with an upvalue name gets an error
>> (e.g. <20>local a; function f() x = {a=1} end<6E>)
(by Edgar Toernig; since 3.1)

** lparser.c
Tue Aug 29 15:56:05 EST 2000
>> error message for `for' uses `while'
(since 4.0a; already corrected)

** lgc.c
Tue Aug 29 15:57:41 EST 2000
>> gc tag method for nil could call line hook
(by ry; since ?)



=================================================================
--- Version 4.0 Beta

** liolib.c
Fri Sep 22 15:12:37 EST 2000
>> `read("*w")' should return nil at EOF
(by roberto; since 4.0b)

** lvm.c
Mon Sep 25 11:47:48 EST 2000
>> lua_gettable does not get key from stack top
(by Philip Yi; since 4.0b)

** lgc.c
Mon Sep 25 11:50:48 EST 2000
>> GC may crash when checking locked C closures
(by Philip Yi; since 4.0b)

** lapi.c
Wed Sep 27 09:50:19 EST 2000
>> lua_tag should return LUA_NOTAG for non-valid indices
(by Paul Hankin; since 4.0b)

** llex.h / llex.c / lparser.c
Wed Sep 27 13:39:45 EST 2000
>> parser overwrites semantic information when looking ahead
>> (e.g. <20>a = {print'foo'}<7D>)
(by Edgar Toernig; since 4.0b, deriving from previous bug)

** liolib.c
Thu Oct 26 10:50:46 EDT 2000
>> in function `read_file', realloc() doesn't free the buffer if it can't
>> allocate new memory
(by Mauro Vezzosi; since 4.0b)



=================================================================
--- Version 4.0

** lparser.c
Wed Nov 29 09:51:44 EDT 2000
>> parser does not accept a `;' after a `return'
(by lhf; since 4.0b)

** liolib.c
Fri Dec 22 15:30:42 EDT 2000
>> when `read' fails it must return nil (and not no value)
(by cassino; since at least 3.1)

** lstring.c/lapi.c
Thu Feb  1 11:55:45 EDT 2001
>> lua_pushuserdata(L, NULL) is buggy
(by Edgar Toernig; since 4.0)

** ldo.c
Fri Feb  2 14:06:40 EDT 2001
>> <20>while 1 dostring[[print('hello\n')]] end<6E> never reclaims memory
(by Andrew Paton; since 4.0b)

** lbaselib.c
Tue Feb  6 11:57:13 EDT 2001
>> ESC (which starts precompiled code) in C is \33, not \27
(by Edgar Toernig and lhf; since 4.0b)

** lparser.c
Tue Jul 10 16:59:18 EST 2001
>> error message for `%a' gave wrong line number
(by Leonardo Constantino; since 4.0)

** lbaselib.c
Fri Dec 21 15:21:05 EDT 2001
>> seg. fault when rawget/rawset get extra arguments
(by Eric Mauger; since 4.0b)

** lvm.c
Wed Jun 19 13:28:20 EST 2002
>> line hook gets wrong `ar'
(by Daniel C. Sinclair; since 4.0.b)

** ldo.c
Wed Jun 19 13:31:49 EST 2002
>> `protectedparser' may run GC, and then collect `filename'
>> (in function `parse_file')
(by Alex Bilyk; since 4.0)




=================================================================
--- Version 5.0 alpha

** lgc.c
Fri Aug 30 13:49:14 EST 2002
>> GC metamethod stored in a weak metatable being collected together with
>> userdata may not be cleared properly
(by Roberto; since 5.0a)

** lapi.c
Thu Nov 21 11:00:00 EST 2002
>> ULONG_MAX>>10 may not fit into an int
(by Jeff Petkau; since 4.0)

** lparser.c
Fri Dec  6 17:06:40 UTC 2002
>> scope of generic for variables is not sound
(by Gavin Wraith; since 5.0a)




=================================================================
--- Version 5.0 beta
** lbaselib.c
Fri Dec 20 09:53:19 UTC 2002
>> `resume' was checking the wrong value for stack overflow
(by Maik Zimmermann; since 5.0b)

** ldo.c
Thu Jan 23 11:29:06 UTC 2003
>> error during garbage collection in luaD_protectedparser is not being
>> protected
(by Benoit Germain; since 5.0a)

** ldo.c (and others)
Fri Feb 28 14:20:33 EST 2003
>> GC metamethod calls could mess C/Lua stack syncronization
(by Roberto; since 5.0b)

** lzio.h/zlio.c
Thu Mar 20 11:40:12 EST 2003
>> zio mixes a 255 as first char in a buffer with EOZ
(by lhf; since 5.0a)



--]]
-----------------------------------------------------------------
-- Lua 5.0 (final)

Bug{
what = [[lua_closethread exists only in the manual]],
report = [[by Nguyen Binh, 28/04/2003]],
patch = [[no patch; the manual is wrong]],
}


Bug{
what = [[attempt to resume a running coroutine crashes Lua]],
example = [[
function co_func (current_co)
   coroutine.resume(co)
end
co = coroutine.create(co_func)
coroutine.resume(co)
coroutine.resume(co)     --> seg. fault
]],
report = [[by Alex Bilyk, 09/05/2003]], 
patch = [[
* ldo.c:
325,326c325
<     if (nargs >= L->top - L->base)
<       luaG_runerror(L, "cannot resume dead coroutine");
---
>     lua_assert(nargs < L->top - L->base);
329c328,329
<   else if (ci->state & CI_YIELD) {  /* inside a yield? */
---
>   else {  /* inside a yield */
>     lua_assert(ci->state & CI_YIELD);
344,345d343
<   else
<     luaG_runerror(L, "cannot resume non-suspended coroutine");
351a350,358
> static int resume_error (lua_State *L, const char *msg) {
>   L->top = L->ci->base;
>   setsvalue2s(L->top, luaS_new(L, msg));
>   incr_top(L);
>   lua_unlock(L);
>   return LUA_ERRRUN;
> }
> 
> 
355a363,368
>   if (L->ci == L->base_ci) {
>     if (nargs >= L->top - L->base)
>       return resume_error(L, "cannot resume dead coroutine");
>   }
>   else if (!(L->ci->state & CI_YIELD))  /* not inside a yield? */
>     return resume_error(L, "cannot resume non-suspended coroutine");
]],
}


Bug{
what = [[file:close cannot be called without a file. (results in seg fault)]],
example = [[
> io.stdin.close()    -- correct call shold be io.stdin:close()
]],
report = [[by Tuomo Valkonen, 27/05/2003]], 
patch = [[
* liolib.c:
161c161
<   if (lua_isnone(L, 1)) {
---
>   if (lua_isnone(L, 1) && lua_type(L, lua_upvalueindex(1)) == LUA_TTABLE) {
]],   --}}
}


Bug{
what = [[C functions also may have stacks larger than current top]],
example = [[
Must recompile lua with a change in lua.c and with lua_assert defined:
* lua.c:
381a382
>   lua_checkstack(l, 1000);
]],
report = [[Alex Bilyk, 09/06/2003]],
patch = [[
* lgc.c:
247c247
<     if (!(ci->state & CI_C) && lim < ci->top)
---
>     if (lim < ci->top)
]],
}

Bug{
what = [[`pc' address is invalidated when a coroutine is suspended]],
example = [[
function g(x)
    coroutine.yield(x)
end

function f (i)
  debug.sethook(print, "l")
  for j=1,1000 do
    g(i+j)
  end
end

co = coroutine.wrap(f)
co(10)
pcall(co)
pcall(co)
]],
report = [[Nick Trout, 07/07/2003]],
patch = [[
* lvm.c:
402d401
<   L->ci->u.l.pc = &pc;
405a405
>   L->ci->u.l.pc = &pc;
676,678c676
<           lua_assert(ci->u.l.pc == &pc &&
<                      ttisfunction(ci->base - 1) &&
<                      (ci->state & CI_SAVEDPC));
---
>           lua_assert(ttisfunction(ci->base - 1) && (ci->state & CI_SAVEDPC));
]]
}


Bug{
what = [[userdata to be collected still counts into new GC threshold,
increasing memory consumption]],
report = [[Roberto, 25/07/2003]],
example = [[
a = newproxy(true)
getmetatable(a).__gc = function () end
for i=1,10000000 do
  newproxy(a)
  if math.mod(i, 10000) == 0 then print(gcinfo()) end
end
]],
patch = [[
* lgc.h:
18c18
< void luaC_separateudata (lua_State *L);
---
> size_t luaC_separateudata (lua_State *L);

* lgc.c:
113c113,114
< void luaC_separateudata (lua_State *L) {
---
> size_t luaC_separateudata (lua_State *L) {
>   size_t deadmem = 0;
127a129
>       deadmem += sizeudata(gcotou(curr)->uv.len);
136a139
>   return deadmem;
390c393
< static void checkSizes (lua_State *L) {
---
> static void checkSizes (lua_State *L, size_t deadmem) {
400c403
<   G(L)->GCthreshold = 2*G(L)->nblocks;  /* new threshold */
---
>   G(L)->GCthreshold = 2*G(L)->nblocks - deadmem;  /* new threshold */
454c457,458
< static void mark (lua_State *L) {
---
> static size_t mark (lua_State *L) {
>   size_t deadmem;
467c471
<   luaC_separateudata(L);  /* separate userdata to be preserved */
---
>   deadmem = luaC_separateudata(L);  /* separate userdata to be preserved */
475a480
>   return deadmem;
480c485
<   mark(L);
---
>   size_t deadmem = mark(L);
482c487
<   checkSizes(L);
---
>   checkSizes(L, deadmem);
]]
}

Bug{
what=[[IBM AS400 (OS400) has sizeof(void *)==16, and a `%p' may generate
up to 60 characters in a `printf'. That causes a buffer overflow in
`tostring'.]],

report = [[David Burgess, 25/08/2003]],

example = [[print{}; (in an AS400 machine)]],

patch = [[
* liolib.c:
178c178
<   char buff[32];
---
>   char buff[128];

* lbaselib.c:
327c327
<   char buff[64];
---
>   char buff[128];
]]
}


Bug{
what = [[syntax `local function' does not increment stack size]],

report = [[Rici Lake, 26/09/2003]],

example = [[
-- must run this with precompiled code
local a,b,c
local function d () end
]],

patch = [[
* lparser.c:
1143a1144
>   FuncState *fs = ls->fs;
1145c1146,1147
<   init_exp(&v, VLOCAL, ls->fs->freereg++);
---
>   init_exp(&v, VLOCAL, fs->freereg);
>   luaK_reserveregs(fs, 1);
1148c1150,1152
<   luaK_storevar(ls->fs, &v, &b);
---
>   luaK_storevar(fs, &v, &b);
>   /* debug information will only see the variable after this point! */
>   getlocvar(fs, fs->nactvar - 1).startpc = fs->pc;
]],

}


Bugs = {

what = [[count hook may be called without being set]],

report = [[Andreas Stenius, 06/10/2003]],

example = [[
set your hooks with

  lua_sethook(L, my_hook, LUA_MASKLINE | LUA_MASKRET, 1);

(It is weird to use a count > 0 without setting the count hook,
but it is not wrong.)
]],

patch = [[
* lvm.c:
69c69
<   if (mask > LUA_MASKLINE) {  /* instruction-hook set? */
---
>   if (mask & LUA_MASKCOUNT) {  /* instruction-hook set? */
]],

}