NEWS

CHANGES IN VERSION 0.20.0
-------------------------

NEW FEATURES

    o rbind() now supports DataFrame objects with the same column names
      but in different order, even when some of the column names are
      duplicated. How rbind() re-aligns the columns of the various objects
      to bind with those of the first object is consistent with what
      base:::rbind.data.frame() does.

    o Add isSequence() low-level helper.

    o Add 'nodup' argument to selectHits().

SIGNIFICANT USER-VISIBLE CHANGES

    o The rownames of a DataFrame are no more required to be unique.

    o Change 'use.names' default from FALSE to TRUE in mcols() getter.

    o Coercion to DataFrame now **always** propagates the names.

    o Rename low-level generic concatenateObjects() -> bindROWS().

    o replaceROWS() now dispatches on 'x' and 'i' instead of 'x' only.

    o Speedup row subsetting of DataFrame with many columns.

DEPRECATED AND DEFUNCT

    o phead(), ptail(), and strsplitAsListOfIntegerVectors() are now defunct
      (after being deprecated in BioC 3.7).

BUG FIXES

    o Fix window() on a DataFrame with data.frame columns.

    o 2 fixes to "rbind" method for DataFrame objects:
      - It now properly handles DataFrame objects with duplicated colnames.
        Note that the new behavior is consistent with base::rbind.data.frame().
      - It now properly handles DataFrame objects with columns that are 1D
        arrays.

    o Fix showAsCell() on nested data-frame-like objects.

    o 2 fixes to "as.data.frame" method for DataFrame objects:
      - It now works if the DataFrame object contains nested data-frame-like
        objects or other complicated S4 objects (as long as these complicated
        objects in turn support as.data.frame()).
      - It now handles 'stringsAsFactors' argument properly. Originally
        reported here: https://github.com/Bioconductor/GenomicRanges/issues/18


CHANGES IN VERSION 0.18.0
-------------------------

NEW FEATURES

    o The package gets a new vignette: S4VectorsOverview.Rnw
      The material in this new vignette comes from the IRangesOverview.Rnw
      vignette located in the IRanges package. All the S4Vectors-specific
      material was moved from the IRangesOverview.Rnw vignette to the new
      S4VectorsOverview.Rnw vignette.

    o All Vector derivatives now support 'x[i, j]' by default. This allows
      the user to conveniently subset the metadata columns thru 'j'.
      Note that GenomicRanges objects have been supporting this feature for
      years but now all Vector derivatives support it. Developers of Vector
      derivatives with a true 2-D semantic (e.g. SummarizedExperiment) need
      to overwrite this.

    o rank() now suports 'by' on Vector derivatives.

    o Add concatenateObjects() generic and methods for LLint, vector, Vector,
      Hits, and Rle objects. This is a low-level generic intended to
      facilitate implementation of c() on vector-like objects.
      The "concatenateObjects" method for Vector objects concatenates the
      objects by concatenating all their parallel slots. The method behaves
      like an endomorphism with respect to its first argument 'x'. Note that
      this method will work out-of-the-box and do the right thing on most
      Vector subclasses as long as parallelSlotNames() reports the names of
      all the parallel slots on objects of the subclass (some Vector subclasses
      might require a "parallelSlotNames" method for this to happen). For those
      Vector subclasses on which concatenateObjects() does not work
      out-of-the-box or does not do the right thing, it is strongly advised
      to override the method for Vector objects rather than trying to override
      the (new) "c" method for Vector objects with a specialized method. The
      specialized "concatenateObjects" method will typically delegate to the
      method below via the use of callNextMethod(). See "concatenateObjects"
      methods for Hits and Rle objects for some examples. No Vector subclass
      should need to override the "c" method for Vector objects.

    o Major refactoring of [[<- for List objects. It's now based on a new
      "setListElement" method for List objects that relies on `[<-` for
      replacement, c() for appending, and `[` for removal, which are the 3
      operations that setListElement() can perform (depending on how it's
      called). As a consequence [[<- now works out-of-the box on any List
      derivative for which `[<-`, c(), and `[` work.

SIGNIFICANT USER-VISIBLE CHANGES

    o endoapply() and mendoapply() are now regular functions instead of
      generic functions.

    o A couple of minor improvements to how default "showAsCell" method
      handles list-like and non-list like objects.

    o Replace strsplitAsListOfIntegerVectors() with toListOfIntegerVectors().
      (The former is still available but deprecated in favor of the latter.)
      The input of toListOfIntegerVectors() now can be a list of raw vectors
      (in addition to be a character vector), in which case it's treated like
      if it was 'sapply(x, rawToChar)'.

    o A couple of optimizations to "[<-" method for DataFrame objects
      (see commit e63f4cfd637e3471e4b04015c2938348df17e14a).

DEPRECATED AND DEFUNCT

    o phead() and ptail() are deprecated in favor of IRanges::heads() and
      IRanges::tails().

    o strsplitAsListOfIntegerVectors() is deprecated in favor of
      toListOfIntegerVectors().

BUG FIXES

    o The mcols() setter no more tries to downgrade to DataFrame a supplied
      right value that extends DataFrame (e.g. DelayedDataFrame).

    o 'DataFrame(I(x)) and as(I(x), "DataFrame")' now drops the I() wrapping
      before storing 'x' in the returned object. This wrapping was ugly, not
      needed, and breaking S4 objects.

    o Fix a couple of long-standing bugs in DataFrame subassignment:
      - Bug in the "[<-" method for DataFrame objects where replacing the
        1st variable with a rectangular object (e.g. x[1] <-
        DataFrame(aa=I(matrix(1:6, ncol=2)))) was returning a DataFrame
        with the "nrows" slot set incorrectly.
      - A couple of bugs in the "replaceROWS" method for DataFrame objects
        when used in "rbind mode" i.e. when max(i) > nrow(x).

    o Fix bug in "cbind" method for DataFrame where it was appending X to
      the column names in some situations (see
      https://github.com/Bioconductor/S4Vectors/issues/8).

    o Fix order() on SortedByQueryHits objects (see
      https://github.com/Bioconductor/S4Vectors/issues/6).

    o Fix bug in internal new_Hits() constructor where it was not returning an
      object of the class specified via 'Class' in some situations.

    o "lapply" for SimpleList objects now calls match.fun(FUN) internally to
      find the function to apply.


CHANGES IN VERSION 0.16.0
-------------------------

NEW FEATURES

    o Introduce FilterResults as generic parent of FilterMatrix.


    o Optimized subsetting of an Rle object by an integer vector. Speed up
      is about 3x or more for big objects with respect to BioC 3.5.

SIGNIFICANT USER-VISIBLE CHANGES

    o coerce,list,DataFrame generates "valid" names when list has none.
      This ends up introducing an inconsistency between DataFrame and
      data.frame but it is arguably a good one. We shouldn't rely on
      DataFrame() to generate variable names from scratch anyway.

BUG FIXES

    o Fix showAsCell() on data-frame-like and array-like objects with a single
      column, and on SplitDataFrameList objects.

    o Calling DataFrame() with explict 'row.names=NULL' should block rownames
      inference.

    o cbind.DataFrame() ensures every argument is a DataFrame, not just first.

    o rbind_mcols() now is robust to missing 'x'.

    o Fix extractROWS() for arrays when subscript is a RangeNSBS.

    o Temporary workaround to make the "union" method for Hits objects work
      even in the presence of another "union" generic in the cache (which is
      the case e.g. if the user loads the lubridate package).

    o A couple of (long-time due) tweaks and fixes to "unlist" method for
      List objects so that it behaves consistently with "unlist" method for
      CompressedList objects.

    o Modify Mini radix C code to accomodate a bug in Apple LLVM version 6.1.0
      optimizer.
      [commit 241150d2b043e8fcf6721005422891baff018586]

    o Fix match,Pairs,Pairs()
      [commit a08c12bf4c31b7304d25122c411d882ec52b360c]

    o Various other minor fixes.


CHANGES IN VERSION 0.14.0
-------------------------

NEW FEATURES

    o Add LLint vectors: similar to ordinary integer vectors (int values at
      the C level) but store "large integers" i.e. long long int values at the
      C level. These are 64-bit on Intel platforms vs 32-bit for int values.
      See ?LLint for more information. This is in preparation for supporting
      long Vector derivatives (planned for BioC 3.6).

    o Default "rank" method for Vector objects now supports the same ties
      method as base::rank() (was only supporting ties methods "first" and
      "min" until now).

    o Support x[[i,j]] on DataFrame objects.

    o Add "transform" methods for DataTable and Vector objects.

SIGNIFICANT USER-VISIBLE CHANGES

    o Rename union classes characterORNULL, vectorORfactor, DataTableORNULL,
      and expressionORfunction -> character_OR_NULL, vector_OR_factor,
      DataTable_OR_NULL, and expression_OR_function, respectively.

    o Remove default "xtfrm" method for Vector objects. Not needed and
      introduced infinite recursion when calling order(), sort() or rank() on
      Vector objects that don't have specific order/sort/rank methods.

DEPRECATED AND DEFUNCT

    o Remove compare() (was defunct in BioC 3.4).

    o Remove elementLengths() (was defunct in BioC 3.4).

BUG FIXES

    o Make showAsCell() robust to nested lists.

    o Fix bug where subsetting a List object 'x' by a list-like subscript was
      not always propagating 'mcols(x)'.


CHANGES IN VERSION 0.12.0
-------------------------

NEW FEATURES

    o Add n-ary "merge" method for Vector objects.

    o "extractROWS" methods for atomic vectors and DataFrame objects now
      support NAs in the subscript. As a consequence a DataFrame can now
      be subsetted by row with a subscript that contains NAs. However that
      will only succeed if all the columns in the DataFrame can also be
      subsetted with a subscript that contains NAs (e.g. it would fail at
      the moment if some columns are Rle's but we have plans to make this
      work in the future).

    o Add "union", "intersect", "setdiff", and "setequal" methods for Vector
      objects.

    o Add coercion from data.table to DataFrame.

    o Add t() S3 methods for Hits and HitsList.

    o Add "c" method for Pairs objects.

    o Add rbind/cbind methods for List, returning a list matrix.

    o aggregate() now supports named aggregator expressions when 'FUN' is
      missing.

SIGNIFICANT USER-VISIBLE CHANGES

    o "c" method for Rle objects handles factor data more gracefully.

    o "eval" method for FilterRules objects now excludes NA results, like
      subset(), instead of failing on NAs.

    o Drop "as.env" method for List objects so that as.env() behaves more like
      as.data.frame() on these objects.

    o Speed up "replaceROWS" method for Vector objects when 'x' has names.

    o Optimize selfmatch for factors.

DOCUMENTATION IMPROVEMENTS

    o Add S4QuickOverview vignette.

DEPRECATED AND DEFUNCT

    o elementLengths() and compare() are now defunct (were deprecated in
      BioC 3.3).

    o Remove "ifelse" methods for Rle objects (were defunct in BioC 3.3),

BUG FIXES

    o Fix bug in showAsCell(x) when 'x' is an AsIs object.

    o DataFrame() avoids NULL names when there are no columns.

    o DataFrame with NULL colnames are now considered invalid.


CHANGES IN VERSION 0.10.0
-------------------------

NEW FEATURES

    o Add SelfHits class, a subclass of Hits for representing objects where the
      left and right nodes are identical.

    o Add utilities isSelfHit() and isRedundantHit() to operate on SelfHits
      objects.

    o Add new Pairs class that couples two parallel vectors.

    o head() and tail() now work on a DataTable object and behave like on an
      ordinary matrix.

    o Add as.matrix.Vector().

    o Add "append" methods for Rle/vector (they promote to Rle).

SIGNIFICANT USER-VISIBLE CHANGES

    o Many changes to the Hits class:
      - Replace the old Hits class (where the hits had to be sorted by query)
        with the SortedByQueryHits class.
      - A new Hits class where the hits can be in any order is re-introduced as
        the parent of the SortedByQueryHits class.
      - The Hits() constructor gets the new 'sort.by.query' argument that is
        FALSE by default. When 'sort.by.query' is set to TRUE, the constructor
        returns a SortedByQueryHits instance instead of a Hits instance.
      - Bidirectional coercion is supported between Hits and SortedByQueryHits.
        When going from Hits to SortedByQueryHits, the hits are sorted by query.
      - Add "c" method for Hits objects.
      - Rename Hits slots:
          queryHits -> from
          subjectHits -> to
          queryLength -> nLnode (nb of left nodes)
          subjectLength -> nRnode (nb of right nodes)
      - Add updateObject() method to update serialized Hits objects from old
        (queryHits/subjectHits) to new (from/to) internal representation.
      - The "show" method for Hits objects now labels columns with from/to by
        default and switches to queryHits/subjectHits labels only when the
        object is a SortedByQueryHits object.
      - New accessors are provided that match the new slot names: from(), to(),
        nLnode(), nRnode(). The old accessors (queryHits(), subjectHits(),
        queryLength(), and subjectLength()) are just aliases for the new
        accessors. Also countQueryHits() and countSubjectHits() are now aliases
        for new countLnodeHits() and countRnodeHits().

    o Transposition of Hits objects now propagates the metadata columns.

    o Rename elementLengths() -> elementNROWS() (the old name was clearly a
      misnomer). For backward compatibility the old name still works but is
      deprecated (now it's just an "alias" for elementNROWS()).

    o Rename compare() -> pcompare(). For backward compatibility the old name
      still works but is just an "alias" for pcompare() and is deprecated.

    o Some refactoring of the Rle() generic and methods:
      - Remove ellipsis from the argument list of the generic.
      - Dispatch on 'values' only.
      - The 'values' and 'lengths' arguments now have explicit default values
        logical(0) and integer(0) respectively.
      - Methods have no more 'check' argument but new low-level (non-exported)
        constructor new_Rle() does and is what should now be used by code that
        needs this feature.

    o Optimize subsetting of an Rle object by an Rle subscript: the subscript
      is no longer decoded (i.e. expanded into an ordinary vector). This
      reduces memory usage and makes the subsetting much faster e.g. it can be
      100x times faster or more if the subscript has many (e.g. thousands) of
      long runs.

    o Modify "replaceROWS" methods so that the replaced elements in 'x' get
      their metadata columns from 'value'. See this thread on bioc-devel:
        https://stat.ethz.ch/pipermail/bioc-devel/2015-November/008319.html

    o Remove ellipsis from the argument list of the "head" and "tail" methods
      for Vector objects.

    o pc() (parallel combine) now returns a List object only if one of the
      supplied objects is a List object, otherwise it returns an ordinary list.

    o The "as.data.frame" method for Vector objects now forwards the
      'row.names' argument.

    o Export the "parallelSlotNames" methods.

DEPRECATED AND DEFUNCT

    o Deprecate elementLengths() in favor of elementNROWS(). New name reflects
      TRUE semantic.

    o Deprecate compare() in favor of pcompare().

    o After being deprecated in BioC 3.2, the "ifelse" methods for Rle objects
      are now defunct.

    o Remove "aggregate" method for vector objects which was an undocumented
      bad idea from the start.

BUG FIXES

    o Fix 2 long-standing bugs in "as.data.frame" method for List objects:
      - must always return an ordinary data.frame (was returning a DataFrame
        when 'use.outer.mcols' was TRUE),
      - when 'x' has names and 'group_name.as.factor' is TRUE, the levels of
        the returned group_name col must be identical to 'unique(names(x))'
        (names of empty list elements in 'x' was not showing up in
        'levels(group_name)').

    o Fix and improve the elementMetadata/mcols setter method for Vector
      objects so that the specific methods for GenomicRanges, GAlignments,
      and GAlignmentPairs objects are not needed anymore and were removed.
      Note that this change also fixes setting the elementMetadata/mcols of a
      SummarizedExperiment object with NULL or an ordinary data frame, which
      was broken until now.

    o Fix bug in match,ANY,Rle method when supplied 'nomatch' is not NA.

    o Fix findMatches() for Rle table.

    o Fix show,DataTable-method to display all rows if <= nhead + ntail + 1


CHANGES IN VERSION 0.4.0
------------------------

NEW FEATURES

    o Add isSorted() and isStrictlySorted() generics, plus some methods.

    o Add low-level wmsg() helper for formatting error/warning messages.

    o Add pc() function for parallel c() of list-like objects.

    o Add coerce,Vector,DataFrame; just adds any mcols as columns on top of the
      coerce,ANY,DataFrame behavior.

    o [[ on a List object now accepts a numeric- or character-Rle of length 1.

    o Add "droplevels" methods for Rle, List, and DataFrame objects.

    o Add table,DataTable and transform,DataTable methods.

    o Add prototype of a better all.equals() for S4 objects.

SIGNIFICANT USER-VISIBLE CHANGES

    o Move Annotated, DataTable, Vector, Hits, Rle, List, SimpleList, and
      DataFrame classes from the IRanges package.

    o Move isConstant(), classNameForDisplay(), and low-level argument
      checking helpers isSingleNumber(), isSingleString(), etc... from the
      IRanges package.

    o Add as.data.frame,List method and remove other inconsistent and not
      needed anymore "as.data.frame" methods for List subclasses.

    o Remove useless and thus probably never used aggregate,DataTable method
      that followed the time-series API.

    o coerce,ANY,List method now propagates the names.

BUG FIXES

    o Fix bug in coercion from list to SimpleList when the list contains
      matrices and arrays.

    o Fix subset() on a zero column DataFrame.

    o Fix rendering of Date/time classes as DataFrame columns.