Data Files

plot reads gnuplot's default text format.

# iteration  continuity  momentum      <- comment line
1  1.0e-1  2.0e-1
2  8.9e-2  ?                           <- missing cell: a gap
3  7.9e-2  1.6e-1   # trailing comment

4  7.1e-2  1.4e-1                      <- one blank line: new block, the line breaks
5  6.3e-2  1.3e-1


1  5.0e-1  9.0e-1                      <- two blank lines: new dataset (index 1)
2  4.0e-1  8.0e-1

Rule	Behaviour
Columns	separated by blanks or tabs
#	starts a comment anywhere on a line; comment-only lines are neither data nor separators
One blank line	ends a block: plotted lines are broken there
Two blank lines	end a dataset, selected with index N (0-based)
?, NaN, inf, non-numbers	missing values: gaps in the plot
Numbers	read as gnuplot (C strtod): the longest leading number counts, 3abc is 3, 1.2.3 is 1.2, 1d3 is 1; hexadecimal too, 0x10 is 16; beyond the real range, a gap
"quoted cell"	one cell, blanks included; not a number
Short rows	the missing columns are missing values
Column 0	the number of the point among those plotted in its dataset, from 0 (after every, as gnuplot)
every I:J:K:L:M:N	points K to M every I of each block, blocks L to N every J of each dataset, from 0

Values are stored flattened with capacities doubled on growth: a large log costs about one real(R8P) per value.

Separators (CSV)

set datafile separator makes other characters separate the cells, as gnuplot:

set datafile separator comma        # CSV; also "," or ";," (any of the characters)
set datafile separator tab          # or "\t"
set datafile separator              # back to whitespace, as unset datafile

iteration,residual,cd                  <- a header of words: a row of missing values
1, 1.0e-1, 0.52                        <- blanks around cells are ignored
2,8.9e-2,                              <- empty cells, also a trailing one, are missing values
3,"7.9e-2",0.49                        <- a quoted cell is read without its quotes

Every separator ends a cell, so two in a row leave an empty one; separators inside double quotes do not count. Comments, blocks and datasets work as above. A header row is a row of missing values, counted by column 0, until an item uses the headers: using 1:"residual", title columnhead or set key autotitle columnhead make the first row of each dataset a header, no longer a point (see Column headers).

Monitoring logs

A solver that appends one line per iteration and flushes it is all --watch needs:

open(newunit=log_unit, file='residuals.dat', status='replace', action='write')
write(log_unit, '(A)') '# iteration  continuity  momentum'
do it = 1, max_it
   ! ... solve ...
   write(log_unit, '(I0,2(1X,ES14.6E3))') it, res_rho, res_mom
   flush(log_unit)
enddo

A cycle can read the last line while the solver is still writing it: a truncated cell becomes a missing value, or a wrong one (7.9 read from a half-written 7.95e-2). The completed line changes the file size again, so the next cycle redraws it correctly.