What changed in EDuke32, build by build

We shouldn't assume a particular bytes-per-line value and use ylookup[] instead. Specifically, windowed modes on Windows use a frame buffer that always has odd x dimension.

- replace bit tweaking for big endian archs with clear code - allow palette lookups >= 128, previously we read into a signed byte

- in Mapster, pre-form the default 10 clip map names before returning from G_CheckCommandLine() so it gets loaded even if we passed no cmdline args. - malloc + strlen + strcpy --> strdup - don't need to spank dead variables ;) - we may call calloc with zero size, which isn't bad by itself, but asserting for non-null afterwards is. Allocs of 0 are implementation-defined, and may well return a null pointer (C99 7.20.3).

Make some computer Duke opponent variables in player.c static.

Gives a couple more fps for scenes where much screen estate is covered by stuff, like when holding the devastator.

With the same setup as before, a screen-filling translucent wall (with nothing drawn behind it) renders at about 7 fps faster (from 60-something fps initially)

A run of consecutive mapstates may share sector/wall/sprite blocks, but the code was deciding whether to free them solely on local properties. Now, save a reference count at the beginning of each such allocated block and free it only if it reaches zero.

Also, always print at least level 5 corruptions (that is, have no instances of completely silent corruptcheck).

This checks for about every possible type of sprite list corruption and reports back with a corruption level 5 if one of the 14 validations fail.

- start scrolling from about half of the visible page - PGUP/DGDN moves the cursor by 1/4 of the page - code cleanup (we only need one displine[80], use clearkeys())

Note the capitalization. I hope that this makes clashes with user variable names less likely.

Also include one comment about a currently failing assertion there.

Previously, if the sprite turned really out to be in void space, either freelist inconsistency (before the list rewrite) or oob access (now) would happen. Also add an bound-checking assert() for insertsprite's sectnum argument (it's not a bound check!)

This might now be even more confusing for users reading both the source and CON code (where the access is necessarily still via '.clipping'), but at least reading the source now makes sense :P

This is so that in the event of a crash, no CON dump will be printed if no (actor or loadactor) CON code was actually executing at the time of the crash, confusing the users. Note: doesn't apply for event code.

We've been running with same-sized actor_t for a while without problems now. In the unlikely event (famous last words) that an issue should pop up later, this revision should be consulted for the differences between the two versions.

The format is snd #<sound number> inst <instance of that sound>: voice <internal voice handle>, ow <owner's sprite ID/-1> (this is the interesting part)

- make a couple variables static in sounds_mapster32.c - KHz --> Hz in initialization text

- make some variables static - conditionally compile out some code with DEBUG_ALLOCACHE_AS_MALLOC - exit(1) on failure (instead of exit(0)) - some formatting

Note: I'm not really sure whether I'm initializing it in the right places.

When drawing from inside the mirror, there may be no 'bunches' (BUILD lingo, not TROR ones) collected, but following code used the first one unconditionally.

Hlines for masked and translucent masked ceiling/floor (sprites). - apply the --> 'do { ... } while (--cnt)' transformation, making these functions iterate cnt+1 times like the asm version. This also fixes an off-by-one issue where sprites or masked ceilings/floors had a one-pixel non-drawn line to the right. - This time, only declare-as-local two 'extern' globals (asm1 and asm2). It seems that I was too eager with "localing" all file-scoped vars earlier. GCC is able to remove the loads from memory inside the loop by itself, whereas clang is not. This is not trivial, since it has to prove that the 'screen' pointer passed to the functions will never alias these globals.

I have to admit, I like its grumbling.

See E4L1 for what this means.

The apparent horizon (in units proportional to pixels by a constant factor) was being calculated without normalizing to viewingrange, resulting a discrepancy between the drawn and actual up/down angle with non-default viewingranges (e.g. with r_usenewaspect=1, or when setting it from setaspect). This also means that the center-of-aim will now be the same regardless of the aspect. This fixes many derived bugs: * inconsistency w/ crosshair when shooting a gun and aiming up/down (obviously) * bugs resulting from being able to look up/down too far (garbage non-slope texture-mapping and viewing BIGORBIT skies above the borders in classic). Note: mods that had workaround logic for this bug will have to remove it...

added: sector extra; viewingrange and yxaspect before the main drawrooms call removed: randomseed

Usage of a search machine by Hendricks.

The replacement of atoi with strtol in r2374 had the side-effect that numbers out of the range of a 32-bit integer were being returned as LONG_MAX/LONG_MIN instead of being converted by taking the bits and re-interpreting them (note that it was a coincidence that atoi behaved that way; to be strict, the behavior was undefined and there is no regression). Now, we implement parsing decimal integers using strtoll (with assuming "long long" being the same as int64_t) and check in which range the number falls. If it's in the range (INT32_MAX <= x <= UINT32_MAX), issue a warning; if it's not in (INT32_MIN <= x <= UINT32_MAX), warn too (though this better ought to be an error?). In each case, the bit representation is converted to the CON number type (int32 to assume maximum portability) by re-interpreting the bits [this is the same as an int/int cast, with GCC at least; more generally, it's implementation-defined per C99].

(Classic renderer). We don't need 64-bit integers; also use the nice asm version of divscale32() if possible.

This fixes mirrors consisting of more than one walls not drawing when the first assigned mirror wall breaks. It still does not let you have more than one mirror (i.e. mirror sectors) showing up at the same time in the scene. Affects classic and Polymost only.

This was causing oob sector accesses with code like this: getwall[hitwall].nextsector temp ifvarvare sector[temp].floorz sector[temp].ceilingz setvar tempb 1 (from DT's HYPERCORE). Also validate quote indices passed to CON's quote and userquote commands at runtime.

Such game text was shown starting from about the center of the screen. The reason for the bug was this code: t += 1 + isdigit(*(t+2)); The sequence points here are at the beginning and end of this assignment expression, and the updating of t may happen anywhere between these (C99 6.5.16 #3). Please don't write such code. When in doubt, and assignment and reference to the same object should be split!

This plays more nicely with automatic formatters. Also indent accordingly.

Because: 1) It used to be like this; 2) guard against bad CON input.

Encountered with Scent.map. Since the SE21 are deleted after they have been used, the search for them in the ST28 code in sector.c may come up empty, in which case sector[-1] would be accessed.

This was discovered because viewscreens (TILE_VIEWSCR) ceased being drawn with the MAXTILES increase (reported by Hank, who also did the bisect).