Invalid data embedded in the QR code

[zepplock]

Hello,

I'm encoding the following string into the QR: http://localhost:4000/v1/sessions/9a454d3bb5d14369842f21a8f9683a87 which results in this image:

The problem is that the actual data embedded in the QR is: http://localhost:4000/v1/sessions/9a454d3bb5d14369842f21a8f9683a8'. Notice the 7 at the end turned into '. I also saw similar errors which had 1 character wrong but not at the end of the string.

Here's my erl -v

erl -v
Erlang/OTP 18 [erts-7.1] [source] [64-bit] [smp:8:8] [async-threads:10] [hipe] [kernel-poll:false] [dtrace]

One thing to note is that when using Q or L error correction modes it is working fine, at least for the given input string.

[zepplock]

Hello Steve, any update on this issue?

[komone]

Looks like it's not the data but the error correction at the root of this. I am thinking currently that the ECC bytes are overriding the data and flipping a bit in the value. Note that $7 = 0x37 and $' = 0x27

[komone]

This doesn't look trivial to get to the root of, but I have a quiet period coming up that may help me focus :)

[zepplock]

Steve, after verifying changes for #8 I found that L Qand H modes are working but now M mode is the only one that fails for a given string (referenced in the original post). Same problem, last character 7 is turned into '

[komone]

Yep the issue in #6 and #8 is different to this issue. There's an error correction issue that's causing a bit to flip in the data (overriding the embedded value). Further examples if you find them could perhaps be helpful.

[zepplock]

Ok, let me see if I can get more strings. Going to be all URL-like strings. I get them 1 in 10 (on average).

[komone]

My current hypothesis is that the ec code is causing a bit flip, so will be interesting to see if other examples could be explained by a single bit flip. I do not yet understand root cause.

[zepplock]

Sent you some more string over email.

[komone]

That's useful test data - thank you, Vlad. I should be able to isolate the issue and resolve it with this test data.

[komone]

Doing a few "syndrome" checks on the ECC shows that there's an error correction code generation problem... close to a solution