Prior to this change, when a cell was clocked, it was always computed with prior values of referred-to cells. This had the effect, for example, of making cells defined only with a base expression as a total of a column, for example, to be out-of-date in that they would take on the value of the total of the prior values of the column, not the total of new current values of the column. This behavior was very counterintuitive.
With this change, updates of cells that have no clocked expression are delayed until after clocked expressions have been recomputed based on prior values and their cells' current values have been updated. Then the computations of those base-only cells use all of the new current values, leaving the spreadsheet in a (maximally) self-consistent state.
Resolves#90.
Co-authored-by: Glen Whitney <glen@studioinfinity.org>
Reviewed-on: #91
Implements row heights and a command-line option `-F nn` for fteapot to set the general font size.
Also improves documentation and uniformizes key bindings somewhat between teapot and fteapot.
Resolves#57.
Co-authored-by: Glen Whitney <glen@studioinfinity.org>
Reviewed-on: #77
This change gets teapot to compile without warnings on OpenSUSE Tumbleweed
with the extended warnings flags that were recently added. It also represents
a more thoroughly (but not exhaustively) tested and debugged version of the
expression-controlled style settings from an earlier commit. The facility
should be ready for use.
It also adds a better pretty-printer for Token values in gdb and extensive
edits to the documentation. It fixes a bug in max/min introduced by the
addition of the Bool type and the switchover to comparisons returning Bool.
It modifies the display of values too long to fit in their cells to show an
initial segment of their contents with a continuation mark rather than
obscuring the contents completely.
Closes#54.
Closes#55.
Closes#44.
Closes#22.
Closes#13.
This macro looks at each cell along some stride through the sheet,
returning the location of the first cell encountered at which a given
expression evaluates to true, or boolean false otherwise.
To make this more useful, this change also revamps the handling of
boolean values, making them much more permissive and coercing to boolean type
in fewer circumstances. It adds the bool() type conversion for explicitly
coercing to boolean type, and fixes a small issue with reading hex-formatted
floating-point values ('p' can be used as an exponent indicator).
Closes#64.
The main new finction is is(), which tests the type of a value. There
are also constants for each of the possible token types, and one new
conversion: number(), which converts a value into the most appropriate number
type. This change also reuses the functions in scanner.h for exracting ints,
floats, and identifiers from strings, rather than either rewriting the
functionality elsewhere, or using the general-purpose scan when only one
token type was of interest.
For clarity, it is valuable for the results of comparisons to show as "true"
and "false" rather than "1" and "0". This change implements a BOOL type
to facilitate that. Since we want logical operators to short-circuit, this
change also obliges the introduction of macros, which are just like functions
except their arguments are not evaluated before control is passed to the
routine implementing the macro.
This change is the "easy" part of #64, but to reduce the amount of duplicated
code, this change also modifies the internals of implementing a function.
The function identifier is now passed to the implementation as well, allowing
many implementations of similar functions to be collapsed more easily.
There turned out to be one blocking technical issue for full adpotion of
tpa format, and that was the fact that the printed representation of
floating values inside the tpa file might not reproduce the same double
when read. This change therefore introduces the "hexact" floating point
format, based on the the %La format string, which produces a hex
representation with exact round trips. While working on this, it was
convenient to add a new representation "compact" which is basically the
shorter of decimal and scientific representations, without trailing zeros.
This is now the default float format, but of course one can select decimal
or scientific formats to restore prior appearance. However, full-precision
compact format is now the (only) format for editing cell contents, as it
is accurate and efficient. Of course you can enter floating point values
in any format you like when typing in a formula.
The addition of several new floating point options overloaded the menus
in terminal teapot, so this change also revamps those menus slightly,
including eliminating the unused MenuChoice struct, and just specifying menus
with an array of strings.
Closes#63.
The primary change is to add a “funcall” token, so that an entire expression
can be encapsulated as a single token. This change is used to allow a cell to
include simply a selection of appropriate semantic tokens. I.e., the content
and iterative content are now each a single token like the value and the
result value. Token vectors are used only as intermediate results in scanning
and parsing.
Not this means the cells are now in effect storing parse trees, so
computation should be slightly faster and future extensions (like #56) should
be facilitated.
This commit also takes the opportunity while internals are being altered to
add another token to a cell for future use for computed attributes, cf #22,
and to change the internal numerical values from double and ints to long
doubles and long longs. However, the change attempts to encapsulate that
choice so it would be easy to change back or change to another representation.
Note that these changes break savexdr(), as the internal binary format of
a cell is now different. Rather than reimplement it, it is deprecated as
the world does not need another binary spreadsheet format. Hence, the
ascii format for teapot spreadsheets becomes the primary file format.
Loading of old xdr files is still supported for backward compatibility.
Closes#59.
Also along the way, various other slight fixes and enhancements crept in,
a partial but probably not exhaustive list of which follows:
Fixes#31.
Further revisions and improvements to documentation.
Make the approximate comparison of floating point values scale more
accurately with the size of the doubles being compared.
Further extensions of absolute and relative cell addressing.
Addition of (circle constant) tau function/constant.
Modified string conversion to simply use internal printing routines, and
to take "scientific" and "decimal" keywords.
Allowed n() function to take a list of values, or just a single location
defaulting to the current location.
Added floor, ceil, trunc, and round functions, and allowed them to be
keywords controlling the int() integer conversion function as well.
Allowed substr() to drop its last argument to go to the end of the string.
Provided an enum of built-in functions to preserve legacy function
identifiers, allowing the large table inside func.c to be reorganized
in a clearer fashion.
Added additional annotation of properties of the built-in functions,
including precedence.
All operators are now also accessible as built-in functions.
Made precedence of unary - lower than ^ to match python.
Avoided inadvertently using FLTK @symbol abbreviations for formulas with
"@" in them.
Prior to this change, every evaluation re-parses the expression. The
intermediate form as a parse tree is never stored anywhere. This change
adds a new token type, FUNCALL, which acts as a node in a parse tree, and
adds an (untested) evaluation method which returns the unevaluated parse tree
as opposed to evaluating as it goes.
This is preparation for a following step in which rather than storing a token
sequence for each variety of content, teapot will store a single token
representing the parse tree, allowing for quicker evaluation when the time
comes; no actual parsing will have to occur on evaluation.
Also a small amount of preparation of storing parse trees instead of
linear sequences of tokens got mixed in here, namely a new type of
token called a "FUNCALL".
This change does away with separate attributes foreground and background,
opting for an array of color "aspects", where the aspect can vary
over an enum which can either be FOREGROUND or BACKGROUND. This will allow
writing a single set of color-editing code which can work either on the
foreground or background color of a cell.
Now palettes are allocated and deallocated in fteapot as well as teapot;
both executables have some color rendering, with fteapot using the cell
foreground and background (there's just no way to set them at the moment).
teapot is so far only using color in the header.
Implements #39.
Now in fteapot you can select any one of the buttons in a modal dialog
with the cursor keys and hit enter and it will be selected.
Fixes#17.
In addition, I clarified that the block attribute options will either set
every cell to bold (say), or unset every cell, or you can cancel to not
actually go through with the block operation.
These changes required adding a custom FLTK dialog, tpt_choose.{h,cxx}.
Along the way, I reflowed fteapot.fl to make it more readable for further
coding.
Finally, there is a small amount of additional prep for color support
that (unfortunately) got mixed in with these changes.
The new function is X(SRC,REF), which briefly for a cell in the neighborhood
of REF returns the corresponding cell in the neighborhood in SRC. For
further details, see the updated documention.
In developing and documenting this function, I refined some of the
existing error messages (e.g. showing the coordinates when there is an
attempt to obtain a cell with a negative coordinate) and improved the
error propagation to increase the chance that the innermost error will
percolate to the top level.
So far, this jst consists of initializing color in curses mode, and making
the display start and end part of intializing and freeing a sheet (so that
it can control allocating the palette, for example, where the data structure
used depends on what kind of display it is.
Next up will be to allocate and destroy the color palette, and set up the
default colors for cells to use (0 for foreground,
TEAPOT_WHITE for background.) The outline beyond that is to allow setting of
the cell colors, then actually display those colors, and finally edit the
palette.
In general, the goal of these changes is to be more type-specific wherever
possible, reduce duplicated code, and avoid magic numbers. The biggest
individual change is to use Location triples of integers wherever possible
rather than three separate int variables. Another very helpful change for
understanding what is going on in the code is introducing the MarkState enum
for tracking the process of marking and unmarking blocks of cell.
The MarkState change was motivated by the issues with Ctrl-C,
and this commit, in fact,
resolves#28.
Because of the lack of a test harness, it is possible that a change of
this scope has created some bugs as well, but teapot was running OK for
me at the time of the commit. However, I don't know how good my coverage of
the changed code was -- I certainly did not save or load a Lotus 1-2-3 file!
Nearly all of the basic operations in eval.c have been implemented for values
of type LOCATION. The behavior is generally modeled after typical vector
arithmetic. For "<", as a particular case, every component must be less
than or equal the corresponding component, and at least one must be strictly
less (<= drops the latter condition, of course). Note "<" returns the number
of strictly less coordinates, but 0 (false) if any coordinate is greater.
Also implements a mechanism (string array Type_Names) for printing type
information and slightly enhanced the error message for type mismatch in +
Closes#16.
