MusicXML import generates unknown errors

• Aug 4, 2015 - 13:54

When importing MusicXML very often it starts with sayïng that the files are damaged, encountering errors. In the details it shows always similar errors sayïng that Measure Number.... are incomplete, Expecting: 29/32; Found: 6091/6720. When ignoring this it shows an apparently correct file.

Importing the same files in MuseScore 1.3 does not create those errors and also not when they are imported in Sibelius or Finale. Tryïng to analyze the mistakes in the files does not give any clue either. I attached 2 examples + with the error windows, (dutch language).

Please have a look.


Agua de Beber. Take a closer look to measure 40 and 42. They are definitely broken.
There is also a weird slur between measure 15 and 16, probably because the pitch of the first note in measure 16 is not correct.

test 12 street rag is a lot more broken. Just check all the short rests all around.

In reply to by Nicolas

Let me just stick to Agua de Beber. If you're refering to the measures 40 and 42 after importing it in MuseScore 2.0.2 I agree with your conclusion, but as said already MuseScore 1.3 and Sibelius don't complain and don't show those "supposed" broken files. See attached pdf results. So for me it is still not yet clear. If you want to make it clear for me you should point it out in the MusicXML file.
That one seems incorrectly interpreted by MuseScore 2.0.2 ?? (similar is true for the other file and many more)

So by ignoring the details and opening it, it shows these details as broken measures and not only that but more is wrong interpreted. However the mistakes are apparently not present in the MusicXML file. For the weird slur(s) I can give an explanation, but that's for me a known and understandable problem.

In reply to by Marc Sabatella

Whatever the reason is if MusScore 1.3 does a good job there is little reason why 2.0.2 makes a mess of it. I know that interpreting MusicXML is a tricky business and since I've converted far over 10.000 files with version 1.3 I know that 1.3 did a marvelous job. The plugin will stop converting and show the mistakes as well, but with exactly pointing to the line in the MusicXML where it goes wrong. The errors as are represented in 2.0.2 really don't give any clue. Then from the over 10.000 files I converted with 1.3 I had to correct globally 50 files, where correction did make sense as well. In other words it looks to me this import function need to be corrected. A first very rough estimation shows that 50% files have a problem opening in 2.0.2. All with the same kind of problem with the measures. So something must be definetly wrong, and I can only use version 1.3 for now. My advice is use the same conversion as in 1.3. (If possible). Or find the hick up.

In reply to by Robipad

In general the 2.0 importer should be *much* more better than 1.3 - probably hundreds of bugs were fixed, many more notations supported, etc. Using the 1.3 importer would be a huge step backwards overall. But there were some significant changes to the parsing between 2.0.1 and 2.0.2 that potentially could have broken a thing or two. It most definitely should *not* be the case that 50% of files will have problems! However, if your files are all from the same source, it is possible that you are simply the same small handful of problems over and over.

If you see other problems that appear to be of a different nature than this particular one, do post them - we can't fix problems that don't get reported.

In reply to by Robipad

Looking further, it appears the problem actually starts in measure 38. It definitely seems corrupted in the original MusicXML file. I can count the beats in the MusicXML and see way more than 4, and you can see this also looking at the imported result in 2.0.2. Still not sure how this relates specifically to what happens in 40 & 42, but it's definitely not valid input as far as I can tell. In measure 40, the input itself seems ok, but the bad tie that started way back in measure 15 is indeed also probably part of the problem.

When I look at it in 1.3, it definitely shows the corruption in measure 38 and the bad tie. It jiust doesn't also report a problem in measure 40. But bad input produces bad results in both versions.

In reply to by Marc Sabatella

You are correct about the measure 38 but not about the tie. I removed the bad ties, that has no influence. I'm still looking for Aqua de Beber with the correct measure 38 (it is somewhere at my disk). The bad ties are a general problem. You are also correct about the same source, because the files I tested come all from PDFtoMusic Pro which converted the BIAB files. Nevertheless I checked also some Wikifonia files and encountered similar mistakes. Additionally I found many more problems in 2.0.2 compared to 1.3. like wrong translation of chords. Practically 2.0.2 has become a big problem for me wrt MusicXML (or MXL).

PDFtoMusic Pro is developed using exported Finale PDF's. Details in PDF's will influence strongly the results. BIAB PDF's did cause them much trouble, but they did a good job to solve a lot. The moment you use another PDF, say from Sibelius or whatever manufacturer new problems will arise. Anybody seems to use its own interpretation but even different pdf drivers from different pdf printers may cause problems I discovered. I will try to gather some new examples for you with less evident mistakes. It seems that Sibelius, Finale and 1.3 are more tolerant to "bad" input.

In reply to by Robipad

This could be, but again, we can't fix problems no one reports, so please post problems you find.

With respect to chords, assuming you mean chord symbols (aka "Harmony" elements), 2.0 should actually recogize *many* more chords than 1.3. Quite literally, *infinitely* more - 1.3 trnalsated only a finite set of chords (which did happen to be copied directly from an older version of BIAB) whereas 2.0 should handle virtually anything that makes syntactic sense at all. If you are encountering a specific chord that is not handled correctly, please post the file in question.

Regaridng the bad ties - there are actually several, it seems. You would probably have to find and remove them all in order to see improvement.

Do you still have an unanswered question? Please log in first to post your question.