chibisi / saucer Goto Github PK

Compiler options should be set using a environmental variable rather than on each compilation function call for practical reasons.

Complete the implementation of the ExternalPtr struct interface with R's externalptr and provide examples in the vignettes (link to readme) and make it available for users in the package.

Write documentation for installing on Fedora OS

Implicit conversion of returned RType variable to SEXP for exported function results in a segfault. For example this

@Export("dot_product") SEXP dot(SEXP x_sexp, SEXP y_sexp)
{
  size_t n = LENGTH(x_sexp);
  assert(n == LENGTH(y_sexp), "Lengths of the input arrays are not equal");
  
  auto x = NumericVector(x_sexp);
  auto y = NumericVector(y_sexp);
  auto result = NumericVector(1L);
  
  for(long i = 0; i < n; ++i)
  {
    result[0] += x[i]*y[i];
  }
  
  return result;
}

gives a segfault, but this:

@Export("dot_product") auto dot(SEXP x_sexp, SEXP y_sexp)
{
  size_t n = LENGTH(x_sexp);
  assert(n == LENGTH(y_sexp), "Lengths of the input arrays are not equal");
  
  auto x = NumericVector(x_sexp);
  auto y = NumericVector(y_sexp);
  auto result = NumericVector(1L);
  
  for(long i = 0; i < n; ++i)
  {
    result[0] += x[i]*y[i];
  }
  
  return result;
}

does not despite the fact that they are both exported.

Currently code gen code in translator is very messy! It uses ~ and ~= to generate code string when it should be using the std.format: format function to do this. This will make the function much more readable and so easier to maintain.

Also move import std.stdio: writeln; in the translator code for R_init_.../R_unload_... functions into the actual functions themselves, rather than outside them.

The following needs to be implemented. $S \equiv Strings, I \equiv Integers$

For $i,j \in I$ and $i < j$

//Row selection
DataFrame df; df.rows[i..j] 

For $i, j \in I$ and $i < j$, and $n, m \in [I, S]$ and if $n, m \in I, n < m$, and if $n, m \in S, n \leq m$
(in column order if string)

//DataFrame slicing
DataFrame[i..j, n..m] //where 

Both should return a new DataFrame. As in the current selection mechanism, if $m \in I$, selection is exclusive of $m$, and if $m \in S$ selection is inclusive of $m$.

Add detailed documentation for the use of the package. Probably a good idea to use GitHub Wiki.

There should be a test script that runs when you create a PR from dev to main (or any suitable action) to ensure that specific constraints are met.

At the moment if an integer array is submitted to a function requiring a NumericVector, we get an type error. This needs to change so that integers submitted in R for NumericVector types in D are automatically converted since int[] -> double[] implicit casting is fine.

1. Remove uses of runtime assert(...) statements and replace them with enforce(...).
2. Change the translator so that exceptions do not cause R to crash and terminate, but to fail and exit gracefully.

There should therefore be only two types of errors:

1. Compilation errors that include static asserts.
2. Runtime errors that fail gracefully back into R's environment with the appropriate error message.

At the moment stringr is the only dependency in the package. It is a great convenience, but not strictly required so remove it. If anything in a dependency changes, it could otherwise change the status of the package.

Complete the implementation of the Environment struct interface with R's environment and ensure its compatibility with Functions. Provide examples in the vignettes (link to readme) and make it available for users in the package.

Finish implementing support for R Function interface and make it available in the API:

• opCall if(...) type filtering for SEXP convertible, also consider double-scoping the static foreach loop.
• scope(exit)/unprotect for generated SEXPs
• Remove alias this implicit casting and review the cast to SEXP and the To!(SEXP) as well as the relevant RType membership
• Set data items to private.
• Play nicely with saucer Environment type (once this is also finished)
• Static member function for creating an R function?

Insert MIT License text into the package

Write documentation for installation the package from scratch, installing R, the compiler, the package, and running the examples:

Ubuntu & Debian Systems - Done
Fedora - Done
Windows
Mac OS

For more information on CRAN policy

1. https://cran.r-project.org/web/packages/policies.html
2. https://cran.r-project.org/web/packages/external_libs.html
3. https://cran.r-project.org/web/packages/submission_checklist.html

Write documentation for installing on Ubuntu OS from scratch

Users need some surety when downloading the package that it is what they expect by verifying the checksum. So release V0.4 should have such a feature to allow this check for users.

At the moment the matrix object supports basic indexing operators $auto \ value = M_{i, j}$ and $M_{i, j} = value$ and so on, it also needs to support full slicing such as $M[i..j, k..l]$ and $M[i..j, k..l] = value$ and $M_{A}[i..j, k..l] = M_{B}$ and so forth. Also, revisit view operations as well as other design and implementation aspects of the RMatrix object.