This is a suite of tests of effects and settings that players tend to implement incorrectly. These are generally geared toward compatibility with Impulse Tracker, and many of these tests are specific to the IT format.
The .wav files included are generated directly from the disk writer for Impulse Tracker v2.14p5, the latest publicly available version. Any player that claims to be IT-compatible should be producing (nearly) the same results. Of course, some variance is acceptable due to different resampling, noise reduction algorithms, etc. — bitwise-exact output is not the goal, identical playback is.
Keep in mind, most of these tests are fairly obscure, and some are highly unlikely to come up in practice, so most players don't get them all correct. Even the "balls-on accurate" XMPlay fails some of these. However, if a player is failing a large portion of them, there's a good chance that it's not going to play "normal" songs acceptably either.
Note that only the first play through the song is relevant. Since IT doesn't reset channel settings when the song loops, some tests that expect the channels to have certain "default" settings (namely #12) will play differently on repeat.
The first step in achieving better playback is to follow the flowchart and effect notes in ITTECH.TXT. Processing everything in the same fashion as IT will help to eliminate many common bugs.
Arpeggio and pitch slide
|01.||Arpeggio and pitch slide||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Some players handle arpeggio incorrectly, storing and manipulating the original pitch of the note instead of modifying the current pitch. While this has no effect with "normal" uses of arpeggio, it causes strange problems when combining arpeggio and a pitch bend.
When this test is played correctly, the first note will portamento upward, arpeggiate for a few rows, and stay at the higher pitch. If this is played incorrectly, it is most likely because the arpeggio effect is not checking the current pitch of the note.
The second note should have a "stepping" effect. Make sure all three notes of the arpeggio are being altered correctly by the volume-column pitch slide, not just the base note. Also, the pitch after all effects should be approximately the same as the third note.
Lastly, note that the arpeggio is set on the first row, prior to the note. Certain players (notably, Modplug) erroneously ignore arpeggio values if no note is playing.
Arpeggio with no value
|02.||Arpeggio with no value||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
(Note: make sure the previous test passes before trying this one.)
If this test plays correctly, both notes will sound the same, bending downward smoothly. Incorrect (but perhaps acceptable, considering the unlikelihood of this combination of pitch bend and a meaningless arpeggio) handling of the arpeggio effect will result in a "stutter" on the second note, but the final pitch should be the same for both notes. Really broken players will mangle the pitch slide completely due to the arpeggio resetting the pitch on every third tick.
Compatible Gxx off
|03.||Compatible Gxx off||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Impulse Tracker links the effect memories for Exx, Fxx, and Gxx together if "Compatible Gxx" in NOT enabled in the file header. In other formats, portamento to note is entirely separate from pitch slide up/down. Several players that claim to be IT-compatible do not check this flag, and always store the last Gxx value separately.
When this test is played correctly, the first note will bend up, down, and back up again, and the final set of notes should only slide partway down. Players which do not correctly handle the Compatible Gxx flag will not perform the final pitch slide in the first part, or will "snap" the final notes.
Compatible Gxx on
|04.||Compatible Gxx on||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
If this test is played correctly, the first note will slide up and back down once, and the final series should play four distinct notes. If the first note slides up again, either (a) the player is testing the flag incorrectly (Gxx memory is only linked if the flag is not set), or (b) the effect memory values are not set to zero at start of playback.
Gxx, fine slides, effect memory
|05.||Gxx, fine slides, effect memory||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
(Note: make sure the Compatible Gxx flag is handled correctly before performing this test.)
EFx and FFx are handled as fine slides even if the effect value was set by a Gxx effect. If this test is played correctly, the pitch should bend downward a full octave, and then up almost one semitone. If the note is bent way upward (and possibly out of control, causing the note to stop), the player is not handling the effect correctly.
Volume column and fine slides
|06.||Volume column and fine slides||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Impulse Tracker's handling of volume column pitch slides along with its normal effect memory is rather odd. While the two do share their effect memory, fine slides are not handled in the volume column.
When this test is played 100% correctly, the note will slide very slightly downward, way up, and then slightly back down.
(Errata: this description previously indicated that the volume column's non-fine slide behavior is "given back" to the effect column.)
* FMOD cuts the sample on the slide up, suggesting that it does implement this behavior but is incorrectly limiting sample pitches.
Note cut with sample
|07.||Note cut with sample||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Some players ignore sample numbers next to a note cut. When handled correctly, this test should play a square wave, cut it, and then play the noise sample.
If this test is not handled correctly, make sure samples are checked regardless of the note's value.
Out-of-range note delays
|08.||Out-of-range note delays||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
This test is to make sure note delay is handled correctly if the delay value is out of range. The correct behavior is to act as if the entire row is empty. Some players ignore the delay value and play the note on the first tick.
Oddly, Impulse Tracker does save the instrument number, even if the delay value is out of range. I'm assuming this is a bug; nevertheless, if a player is going to claim 100% IT compatibility, it needs to copy the bugs as well.
When played correctly, this should play the first three notes using the square wave sample, with equal time between the start and end of each note, and the last note should be played with the noise sample.
Sample change with no note
|09.||Sample change with no note||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
If a sample number is given without a note, Impulse Tracker will play the old note with the new sample. This test should play the same beat twice, exactly the same way both times. Players which do not handle sample changes correctly will produce various interesting (but nonetheless incorrect!) results for the second measure.
The pattern loop effect is quite complicated to handle, especially when dealing with multiple channels. Possibly the most important fact to realize is that no channel's pattern loop affects any other channel — each loop's processing should be entirely contained to the channel the effect is in.
Another trouble spot that some players have is dealing correctly with strange situations such as two consecutive loopback effects, e.g.:
000 | ... .. .. SB0 001 | ... .. .. SB1 002 | ... .. .. SB1 003 | ... .. .. .00
To prevent this from triggering an infinite loop, Impulse Tracker sets the loopback point to the next row after the last SBx effect. This, the player flow for the above fragment should be 0, 1, 0, 1, 2, 2, 3.
Another point to notice is when a loop should finish if two channels have SBx effects on the same row:
000 | ... .. .. SB0 | ... .. .. SB0 001 | ... .. .. SB1 | ... .. .. SB2
What should happen here is the rows continue to loop as long as ANY of the loopback counters are nonzero, or in other words, the least common multiple of the total loop counts for each channel. In this case, the rows will play six times — two loops for the first channel, and three for the second.
When correctly played, this test should produce a drum beat, slightly syncopated. The entire riff repeats four times, and should sound the same all four times.
Infinite loop exploit
|11.||Infinite loop exploit||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
(Note: on this test, "fail" status is given for players which deadlock while loading or calculating the duration, and is not based on actual playback behavior. Incidentally, this will cause Impulse Tracker to freeze.)
This is a particularly evil pattern loop setup that exploits two possible problems at the same time, and it will very likely cause any player to get "stuck".
The first problem here is the duplicated loopback effect on the first channel; the correct way to handle this is discussed in the previous test. The second problem, and quite a bit more difficult to handle, is the seemingly strange behavior after the third channel's loop plays once. What happens is the second SB1 in the first channel "empties" its loopback counter, and when it reaches the first SB1 again, the value is reset to 1. However, the second channel hasn't looped yet, so playback returns to the first row. The next time around, the second channel is done, but the first one needs to loop again — creating an infinite loop situation. Even Impulse Tracker gets snagged by this.
Tremor is probably one of the least predictably implemented effects. Every player seems to have a different idea of what it's supposed to do. Rather than try to explain every player's implementation and risk confusing things even further, we'll simply stick with describing what Impulse Tracker does.
Like many other effects, tremor has an effect memory. This memory stores both the "on" AND "off" values at once — they are not saved independently.
Also, when tremor is given a zero on- or off-value, that value is handled as one tick instead. That is, I03 plays the same way as I13: play for one tick, off for three.
Another potential snag is what happens when the note is "off" when the row changes and the new row does not have a tremor effect. In this case, the volume is always restored to normal, and the next time the effect is used, the off-tick counter picks up right where it left off.
Finally, the only time the current tremor counts are reset is when the playback is interrupted. Otherwise, the only part of the player code that should even touch these values is the tremor effect handler, and it only ever decreases the values... well, until they hit zero, at that point they are obviously reset; but also note, the reset is independent for the on-tick and off-tick counters.
When this test is played correctly, both notes should play at exactly the same intervals.
Tremor with old effects
|13.||Tremor with old effects||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Just when you think you've figured out tremor, guess what – it's even more annoying. With Old Effects enabled, all non-zero values are incremented, so I40 with old effects means play for five ticks, and turn off for one; I51 means play for six and off for two.
When this test is played correctly, both notes should play at exactly the same intervals.
Ping-pong loop and sample number
|14.||Ping-pong loop and sample number||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
A lone sample number should not reset ping pong loop direction.
In this test, the sample should loop back and forth in both cases. If the sample gets "stuck", check that the player is not touching the loop direction unnecessarily.
The retrig effect, in theory, should create an internal counter that decrements from the "speed" value (the 'y' in Qxy) until it reaches zero, at which point it resets. This timer should be unaffected by changes in the retrig speed – that is, beginning a retrig and then changing the speed prior to the next retrig point should not affect the timing of the next note. Additionally, retrig is entirely independent of song speed, and the counter should reset when a new note is played.
As a side note, I would like to point out that the bassdrum sample uses a silent loop at the end. This is a workaround for Impulse Tracker's behavior of ignoring the retrig effect if no note is currently playing in the channel. Some people seem to have misinterpreted this, coming to the conclusion that retrig values greater than the song speed are ignored. However, this behavior is rather inconvenient when dealing with very short samples. I encourage the authors of other players to treat this behavior as a bug in Impulse Tracker's playback engine and retrigger notes when the timer expires regardless of the current state of the channel.
* Modplug both loads the song incorrectly due to ignoring the initial tempo of 31, and also mishandles the effect.
This test checks the overall handling of the global volume (Vxx) and global volume slide (Wxx) effects. If played properly, both notes should fade in from silence to full volume, back to silence, and then fade in and out a second time. (Note that the volume should be fading to and from maximum, i.e. two W80 effects at speed 9 should change the volume from 0 to 128.)
If the notes start at full volume instead of fading in, the V80 in channel 2 is probably overriding the V00 in channel 3 on the first row. Similarly, if the volume is suddenly cut on row 4, the V00 is probably incorrectly taking precedence over the V80. Generally, for effects that alter global state, the highest-numbered channel takes precedence.
Since two W80 effects at speed 9 raise the volume from zero to the maximum of 128, the V80 on row 3 should not have any effect on the volume.
Also, there are two spurious volume effects in channel 3, on rows 7 and 11. Both of these effects should be ignored as out-of-range data, not clamped to the maximum (or minimum!) volume.
Finally, the previous value for the global volume slide effect is saved per channel. If the volume fades in and out, and does not fade back in, it is almost certainly because separate global volume slide parameters are not stored for each channel.
Pattern row delay
|17.||Pattern row delay||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Impulse Tracker has a slightly idiosyncratic way of handling the row delay and note delay effects, and it's not very clearly explained in ITTECH.TXT. In fact, row delay is mentioned briefly, albeit in a rather obscure manner, and it's easy to glance over it. Considering this, it comes as no surprise that this is one of the most strangely and hackishly implemented behaviors, so much that pretty much everyone has written it off as an IT replayer bug.
In fact, it's not really a bug, just another artifact of the way Impulse Tracker works. See, row delay doesn't touch the tick counter at all; it sets its own counter, which the outer loop uses to repeat the row without processing notes. The difference seems obscure until you try to replicate the retriggering behavior of SEx on the same row with SDx. SDx, of course, operates on all ticks except the first, and SEx repeats a row without handling the first tick, so the notes and first-tick effects only happen once, while delayed notes play over and over as many times as the SEx effect says.
Sound weird? Not really. As I said, it's even mentioned in ITTECH.TXT. Check the flowchart under "Effect Info", and look at the words "Row counter". This particular variable just says how many times the row plays (without replaying notes) before moving to the next row. That's all. You just play the row more than once.
The other quirk is how to handle more than one SEx value on the same row. This is a bit weird, but also straightforward; the first SEx always takes precedence, no matter what. SE0 means play the row only once, SE1 means play it twice, etc.
This test plays sets of alternating bassdrum and snare, followed by a short drum loop. In the first part, the bassdrum should play 2, 1, 7, 5, 1, and 5 times before each respective snare drum.
Sample number causes new note
|18.||Sample number causes new note||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
A sample number encountered when no note is playing should restart the last sample played, and at the same pitch. Additionally, a note cut should clear the channel's state, thereby disabling this behavior.
This song should play six notes, with the last three an octave higher than the first three.
The random waveform for vibrato/tremolo/panbrello should not use a static lookup table, but should be based on some form of random number generator. Particularly, each playing channel should have a different value sequence.
Correct playback of this song should result in three stereo effects. It might also be helpful to view the internal player variables in Impulse/Schism Tracker's info page detail view (page up from the sample VU meters).
* Apparently OpenCP doesn't implement Yxx.
Pan swing and set panning effect
|20.||Pan swing and set panning effect||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
A couple of brief notes about instrument pan swing: All of the values are calculated with a range of 0-64. Values out of the 0-64 range are clipped. * The swing simply defines the amount of variance from the current panning value.
Given all of this, a pan swing value of 16 with a center-panned (32) instrument should produce values between 16 and 48; a swing of 32 with full right panning (64) will produce values between 0 -- technically -32 -- and 32.
However, when a set panning effect is used along with a note, it should override the pan swing for that note.
This test should play sets of notes with: Hard left panning Left-biased random panning Hard right panning Right-biased random panning Center panning with no swing Completely random values
Pitch slide limits
|21.||Pitch slide limits||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Impulse Tracker always increases (or decreases, of course) the pitch of a note with a pitch slide, with no limit on either the pitch of the note or the amount of increment.
An odd side effect of this test is the harmonic strangeness resulting from playing frequencies well above the Nyquist frequency. Different players will seem to play the notes at wildly different pitches depending on the interpolation algorithms and resampling rates used. Even changing the mixer driver in Impulse Tracker will result in different apparent playback. The important part of the behavior (and about the only thing that's fully consistent) is that the frequency is changed at each step.
Zero value for note cut and note delay
|22.||Zero value for note cut and note delay||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Impulse Tracker handles SD0 and SC0 as SD1 and SC1, respectively. (As a side note, Scream Tracker 3 ignores notes with SD0 completely, and doesn't cut notes at all with SC0.)
If these effects are handled correctly, the notes on the first row should trigger simultaneously; the next pair of notes should not; and the final two sets should both play identically and cut the notes after playing for one tick.
Portamento with no note
|23.||Portamento with no note||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Stray portamento effects with no target note should do nothing. Relatedly, a portamento should clear the target note when it is reached.
The first section of this test should first play the same increasing tone three times, with the last GFF effect not resetting the note to the base frequency; the next part should play two rising tones at different pitches, and finish an octave lower than it started.
Short envelope loops
|24.||Short envelope loops||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
Envelope loops should include both of the loop points. Each instrument in this test should play differently: first with no envelope, then a stuttering effect, a rapid left/right pan shift, and finally an arpeggio effect.
Portamento and pitch slide
|25.||Portamento and pitch slide||schism||chibi||openmpt||libmodplug||xmplay||bassmod||dumb||ocp||mikmod||mikit||xmp||fmod|
The first two segments of this test should both play identically, with a slight downward slide followed by a faster slide back up, and then remain at the starting pitch. The third and fourth are provided primarily for reference for possible misimplementations. (The third slides down and then continually rises; the fourth slides down slightly and then plays a constant pitch.)
If the first and third segment are identical, then either Fx and Gxx aren't sharing their values, or the Fx values aren't being multiplied (volume column F1 should be equivalent to a "normal" F04).
If the first and fourth are identical, then the volume column is most likely being processed prior to the effect column. Apparently, Impulse Tracker handles the volume column effects last; note how the G04 needs to be placed on the next row after the F1 in the second segment in order to mirror the behavior of the first.
Application versions tested
For reference, the above results were tested with:
- Schism Tracker 20090728
- Chibi Tracker 1.3
- OpenMPT 1.18.02.00
- libmodplug 0.8.7 (Linux GStreamer plugin)
- XMPlay 3.5
- BASSMOD 2.0-Linux
- DUMB 0.9.3
- Open Cubic Player 0.1.17
- MikMod 3.2.2-beta1
- MikIT 1.00-Linux
- XMP 3.2.0
- FMOD Ex 4.24.07
- 20050522: First release
- 20050525: Added test #12
- 20050527: Added test #13
- 20090516: Added tests #14 and #15
- 20090613: Added test #16
- 20090618: Added test #17
- 20090711: Removed test #13 (everything plays arpeggios right, don't remember why I added this); changed #15 to check Q00 with no prior value; and added tests #18 through #24
- 20090720: Modified #6 to do what it was supposed to do in the first place; made #23 somewhat more thorough; replaced #13 with a new test; added #25
- 20090801: Modified #1 to check that arpeggio is still set if no note was playing; added more stuff to #3 and #4; replaced #13 again.