- Change locale from de_DE to fr_FR.utf8
- Expect correct thousands separator
This also partly undoes the previous commit (it removes the
test for gcc 5 or more)
Apparently gcc V4.8 provides a defective implementation of
std::locale which first doesn't support std::numpunct and second
throws the wrong exception type.
Unfortunately boost::locale isn't a solution because it uses the
gcc facets.
So for now, we don't compile that bit if gcc is too old.
To allow for a difference of 1 on very large numbers (>1e16). At the
same time made the test harder with a wider range, skipping cases where
the result overflows. Use GncRational::valid() instead of testing
components for both overflow and NaN.
Loses three bits so GncInt128 becomes really a GncInt125, but we don’t
really need the single order-of-magnitude: 10**38 is big enough. Saves
a full word of memory for each GncInt128, which means 2 words for GncRational.
That’s a 33% saving in memory for 64-bit and makes the object size the
same (32 bytes) for all architectures.
More consistent with GncNumeric and saves a word of memory per instance.
Still bleeping huge because the two GncInt128s each need 128 bits (2 or 4 words)
plus a word for status (for 3 bits!).
Also provide a couple of convenience functions, is_big() and valid() to
test if the either numerator and denominator is big or overflowed or NaN.
Similar to GncRational, except that it’s based on int64_t instead of
GncInt128 and throws instead of using a status byte.
Most calculations are performed using GncRational, the result is then
rounded (RoundType::half_down) to fit. GncRational should be used in
circumstances where the automatic rounding is undesirable.
- Adjust description on the assistant's start page to match current behavior
- Hide separator buttons in case the file format is set to fixed width
- For fixed width display some instructions on how to manipulate columns
- Move the error messages to below the preview table
This happens because in case std::string is the base type for the tokenizer
the offsets are interpreted as byte offsets although some characters
may consist of more that one byte in utf-8. This is not so for std::wstring
so to solve this issue the fixed width tokenizer is changed to use
wide character strings internally.
Each column change would cause the code to completely delete the
existing columns and then add new ones from scratch. For some
reason his upsets the MOVE_NOTIFY event, which is triggered
whenever you move the mouse pointer around. I suspect it internally
keeps a link to the last hovered column to check if the move action
moves the mouse to a different column. As the code removed the columns
this internal references becomes invalid.
I have worked around this by no longer completely recreating the
columns. Instead columns are added or removed as needed to follow
the changes in the data model.
At the same time the refresh function has been split into several
smaller ones.
