Show Posts
1
Miscellaneous / Re: [Legion] WDB5 format
« on: May 04, 2016, 09:01:19 pm »
Yeah, basically old school DBC column guessing strategy.
I use a priority-based system, like:
1. If every value in the column is 0, it's an integer.
2. If every value in the column exists in our string position map, it's a string.
3. If every value in the column where it is not 0 has the exponent bits for floats set to a number greater than ~120, it's a float.
4. Else, it's an integer.
A couple notes for WDB5: string and float fields still -must- be 4 bytes. So if the field is less than 4 bytes, it is automatically an integer.
Also, there are going to be a lot of problems with signed and unsigned integers if you autodetect like this. Foreign keys are generally unsigned and it sometimes makes a big difference. Other values are generally signed, and it sometimes makes a big difference. You might be able to write some form of autodetection for signed vs unsigned (or take it from somewhere else) but it is going to be annoyingly complicated.
Lastly, I recommend against a dedicated flags column type. At the end of the day, displaying the numbers as 0x(some hex value) doesn't actually make flags significantly more clear or useful (imo), so it's best to just parse flags you want later, after converting to CSV or some other form.
I use a priority-based system, like:
1. If every value in the column is 0, it's an integer.
2. If every value in the column exists in our string position map, it's a string.
3. If every value in the column where it is not 0 has the exponent bits for floats set to a number greater than ~120, it's a float.
4. Else, it's an integer.
A couple notes for WDB5: string and float fields still -must- be 4 bytes. So if the field is less than 4 bytes, it is automatically an integer.
Also, there are going to be a lot of problems with signed and unsigned integers if you autodetect like this. Foreign keys are generally unsigned and it sometimes makes a big difference. Other values are generally signed, and it sometimes makes a big difference. You might be able to write some form of autodetection for signed vs unsigned (or take it from somewhere else) but it is going to be annoyingly complicated.
Lastly, I recommend against a dedicated flags column type. At the end of the day, displaying the numbers as 0x(some hex value) doesn't actually make flags significantly more clear or useful (imo), so it's best to just parse flags you want later, after converting to CSV or some other form.