duncantl / codeanalysis Goto Github PK

The code for for loops with assignments assumes that a brace contains the body of a loop. This should be a straightforward fix, I'm just opening this so that I remember to do it. While I'm at it I should put the documentation in the proper place.

# with brace is fine
> forLoopToLapply(quote(for(i in 1:5){ z[[i]] = i}))                                              
z[1:5] <- lapply(1:5, function(i) {
    i
})

> forLoopToLapply(quote(for(i in 1:5) z[[i]] = i))
Error in expr[[1]] : object of type 'symbol' is not subsettable

Enter a frame number, or 0 to exit   

1: forLoopToLapply(quote(for (i in 1:5) z[[i]] = i))
2: forLoopWithUpdates(forloop, deps)
3: isSimpleIndexAssign(lastline, global_update, ivar)

# no assignment, no brace works
> forLoopToLapply(quote(for(i in 1:5) print(i)))
lapply(1:5, function(i) print(i))

CodeAnalysis::getDepends() seems to miss quite a few statements, such as write.table(), ggsave(), or could not gather the filename properly if it's not specified explicitly by file = value.

For example, this file test_getDepends_example_script.R with the following content

# pardon the syntax, it's for static analysis testing only
X = data.frame('name' = c('a', 'b', 'c'), 'value' = c(1:3))

read.table('file_1.csv')
read.table(file = 'file_2.csv')
read.csv('file_3.csv')
read.csv(file = 'file_4.csv')
write.table('file_5.csv')
write.table(file = 'file_6.csv')
write.csv(X, 'file_7.csv')
write.csv(X, file = 'file_8.csv')

p = ggplot2::ggplot(X) +
  geom_point(aes(x=name,y=value))

ggplot2::ggsave(filename = 'a_plot.png', plot = p)

when tested with CodeAnalysis::getDepends('test_getDepends_example_script.R'), will result in the following

    filename  operation expressionNum
1 file_1.csv read.table             2
2 file_2.csv read.table             3
3 file_3.csv   read.csv             4
4 file_4.csv   read.csv             5
5       <NA>  write.csv             8
6 file_8.csv  write.csv             9

---------- The following code will create the necessary files to reproduce the issue ----------

cat > test_getDepends_example_script.R <<EOL
# pardon the syntax, it's for static analysis testing only
X = data.frame('name' = c('a', 'b', 'c'), 'value' = c(1:3))

read.table('file_1.csv')
read.table(file = 'file_2.csv')
read.csv('file_3.csv')
read.csv(file = 'file_4.csv')
write.table('file_5.csv')
write.table(file = 'file_6.csv')
write.csv(X, 'file_7.csv')
write.csv(X, file = 'file_8.csv')

p = ggplot2::ggplot(X) +
  geom_point(aes(x=name,y=value))

ggplot2::ggsave(filename = 'a_plot.png', plot = p)
EOL

cat > test_getDepends.R <<EOL
CodeAnalysis::getDepends('test_getDepends_example_script.R')
EOL

I've been using the recent changes to rstatic that @nick-ulle has made, and I'd like to make sure CodeAnalysis has the correct version of rstatic.

I added this line to the DESCRIPTION:

Imports: CodeDepends, rstatic (>= 0.0.0.9052)

This versioning doesn't seem to be enforced when I run R CMD INSTALL from the command line, when I load the package with library, or when I call the functions that use rstatic::. That is, I change the line to require rstatic (>= 1.0), which doesn't exist, and R never complains that I don't have that version.

This happens with R 3.5.3 on my local Mac, and R 3.6.0 on Linux.

The R Extensions manual has a section on package dependencies that says:

Version requirements can be specified and are checked when the namespace is loaded (since R >= 3.0.0).

What am I doing wrong?