Oracle's ( or not) weekday convention - sql

Question

I have written a query for my co-workers to use, that involves pulling data from Oracle Express g11 on a given date.
The date that query uses is last weekday(yesterday or Friday).
For some reason same query returns different number to represent the day of the week.
Example:
SELECT TRUNC('13June2016') - TRUNC('13June2016', 'D')
FROM DUAL
When run on my computer this query returns 0 but on my co-worker's computer it returns 1. This happens on every SQL client/interface available to us.
To the best of my knowledge our computers and users in DB are setup the same but clearly some setting is off.
Frankly I tried to google where does this discrepancy comes from but I received a lot of unrelated results. I'm lost because I'd don't even know where to look for the setting(or bug not sure) that accounts for this difference.

Answer 1

The D format modifier relies on the session's NLS settings:
alter session set nls_territory = 'America';
alter session set nls_date_format = 'YYYY-MM-DD';
select to_char(date '2016-06-13', 'D') char_d,
trunc(date '2016-06-13', 'D') trunc_d,
date '2016-06-13' - trunc(date '2016-06-13', 'D') diff1,
trunc(sysdate) - trunc(sysdate, 'D') diff2
from dual;
C TRUNC_D DIFF1 DIFF2
- ---------- ---------- ----------
2 2016-06-12 1 1
alter session set nls_territory = 'United Kingdom';
alter session set nls_date_format = 'YYYY-MM-DD';
select to_char(date '2016-06-13', 'D') char_d,
trunc(date '2016-06-13', 'D') trunc_d,
date '2016-06-13' - trunc(date '2016-06-13', 'D') diff1,
trunc(sysdate) - trunc(sysdate, 'D') diff2
from dual;
C TRUNC_D DIFF1 DIFF2
- ---------- ---------- ----------
1 2016-06-13 0 0
(Interestingly in producing that I stumbled over bug 14073795 when using trunc(date '2016-06-13'); the trunc is redundant and it seems quite inconsistent - changing a column alias can make it appear or disappear - so I don't think it's relevant to what you're seeing).
So it seems that you and your colleague have your PCs in different locales, and the clients you tested in either inherited the locale or explicitly set the territory differently.
To get consistent results regardless of the locale/NLS settings you can use the 'IW' format element instead of 'D', as that gives:
Same day of the week as the first day of the calendar week as defined by the ISO 8601 standard, which is Monday
alter session set nls_territory = 'America';
alter session set nls_date_format = 'YYYY-MM-DD';
select to_char(date '2016-06-13', 'IW') char_iw,
trunc(date '2016-06-13', 'IW') trunc_iw,
date '2016-06-13' - trunc(date '2016-06-13', 'IW') diff1,
trunc(sysdate) - trunc(sysdate, 'IW') diff2
from dual;
CH TRUNC_IW DIFF1 DIFF2
-- ---------- ---------- ----------
24 2016-06-13 0 0
alter session set nls_territory = 'United Kingdom';
alter session set nls_date_format = 'YYYY-MM-DD';
select to_char(date '2016-06-13', 'IW') char_iw,
trunc(date '2016-06-13', 'IW') trunc_iw,
date '2016-06-13' - trunc(date '2016-06-13', 'IW') diff1,
trunc(sysdate) - trunc(sysdate, 'IW') diff2
from dual;
CH TRUNC_IW DIFF1 DIFF2
-- ---------- ---------- ----------
24 2016-06-13 0 0
You can read more about how the format models are used in trunc() and round() functions, and more generally about date format models.
Relying on implicit conversion is also not a good idea; TRUNC('13June2016', 'D') is implicitly converting the string '13June2016' to a date using your session's NLS_DATE_FORMAT setting. Running that in a session with a different setting will error. And TRUNC('13June2016') is redundant as the implicitly-converted date will already have its time set to midnight, if it doesn't error.
You should always either explicitly convert the string and specify the format model, e.g. TRUNC(TO_DATE('13June2016', 'DDMonthYYYY', 'NLS_DATE_LANGUAGE=ENGLISH'), 'D') (where the third argument is needed in case the session date language is something else, which would stop June being recognised; or more simply use an ISO date literal like DATE '2016-06-13'. You may be using a datetime variable in your real code of course.