hhoeflin / hdf5r Goto Github PK

When accessing an array with

dset[, c(2,1)]

it results in an error. The underlying reason is that a hyperslab with a negative stride is internally requested, which is not supported by HDF5. The stride-length has to always be non-negative.

Fix windows build for the following use cases:

• Support CRAN windows build (e.g. on https://win-builder.r-project.org)
• devtools::install_github(hhoeflin/hdf5r)

It can happen that a user wants to close all relevant open HDF5 IDs. A function to simplify this process is missing.

insert check before expand f <- function() dset2[10:11, 9] <- matrix(rep(0, 2*9), nrow = 2)

Currently in read and write operations, various R6 objects are being created. This R6 object creation takes a few milliseconds. On large read and writes, this isn't a problem. On small read/writes, the performance overhead is quite substantial.

Solution: Avoid the creation of R6 objects and do the necessary bookkeeping internally.

Copied from the h5 Makefile:

check-asan-gcc: $(PKG_NAME)_$(PKG_VERSION).tar.gz
	@boot2docker up
	$(shell boot2docker shellinit)
	@docker run -v "$(CURRENT_DIR):/mnt" mannau/r-devel-san /bin/bash -c \
		"cd /mnt; apt-get update; apt-get clean; apt-get install -y libhdf5-dev; \
		R -e \"install.packages(c('Rcpp', 'testthat', 'roxygen2', 'highlight', 'zoo', 'microbenchmark'))\"; \
		R CMD check $(PKG_NAME)_$(PKG_VERSION).tar.gz; \
		cat /mnt/h5.Rcheck/00install.out"

check-valgrind: $(PKG_NAME)_$(PKG_VERSION).tar.gz 
	@rm -rf $(CHECKPATH)
	$(R) CMD check --no-clean --use-valgrind $(PKG_NAME)_$(PKG_VERSION).tar.gz

Want functionality to see all open files and reuse R6 objects related to an open file

When accessing dimensions using a logical vector, it is not appropriately translated to a numeric vector

We should check the NOTE for Non-FOSS package license.
Is there a reason why the Apache license is used or could be also use BSD_2_clause + file LICENSE?

Windows test cases fail

Reading entire dataset using [] (instead of [,]) for 2-dimensional objects should also work, like in matrix(1:9, nrow = 3)[]

Use pkdown to create package documentation.

Reading and writing of data-frames has a bug when a reshuffle is needed

Long usage lines do not get broken at the end of the line appropriately

The readme could be more informative

At several places it is out of date or pointing to an incorrect repo

In R, the dimensions of each dataset are always written in reverse order (to accomodate the fact that In R the first dimension changes fastest, but in C and hdf5, the last dimension changes fastest).

When printing dataset dimensions using ls, the hdf5 ordering is written out, not the reverse order

On appveyor, a note is mentioned that R_RegisterRoutines is called.

Ensure that the note is removed.

Provide options to specify dimension (if maxdim is needed, should specify a space).

How to set up dataset with specified maximum dimensions, only using set_extent?

This should work without setting chunk size: dset1 <- createDataSet(file, "testmat_1", testmat_n, space = dspace)

Implement test cases DataSet-Select-Elem/DataSet-Select-Hyperslab needs refactoring of H5D$read functions (separate dataspace from read).

No Support for on-the-fly group creation / Recursive group creation ? e.g. file[["testgroup/testset"]] <- 1:10 without existing testgroup

Need to check that the built-in browser based help is working properly

Need to go through the manual and check for any issues.

Getting a warning message at the end of the check that 'cleanup' is deprecated

H5Class_overview is currently not working since inst/manual/function_overview.html is missing:

> H5Class_overview()
Error in browseURL(system.file("manual/function_overview.html", package = "hdf5r"),  : 
'url' must be a non-empty character string

We should either:

• Add the manual file inst/manual/function_overview.html
• Delete H5Class_overview and all references to it

Currently, the evaluate_arguments function that is a helper function for subset reading and assignment is relatively slow and should be improved

some common_functions seem to be assigned to some classes where they can't always give a result