Skip to content

Round trip between package_version and character

5 messages · Lluís Revilla, Henrik Bengtsson, Josiah Parry +1 more

Original
Lluís Revilla
# 
Hi all,

I was trying to download the source code of a package using the versioned
urls (https://CRAN.R-project.org/package=%s&version=%s). I used
packageVersion to get the version number but it failed because it uses the
function pacakge_version that replaces the "-" the version number had.

Indeed, the examples show that package_version("1.2-4") represents it as
"1.2.4". However, when converted back to characters (via as.character or
similar approaches) it keeps the "." instead of reverting back to "-". The
work around is essay enough: packageVersion(pkg, field = "Version") so my
code is fixed but it might not be obvious in other cases.

Could the round trip revert to the original package version provided by the
maintainer ( so that identical(as.character(package_version("1.2-4")),
"1.2-4") is TRUE) ?
This would simplify manipulations of package versions (currently ~2000
packages on CRAN use "-" on their version).

Cheers,

Llu?s Revilla

PS: The description at the top of the help page ?download.packages doesn't
fully match the content. I think it refers to the functions documented on
?update.packages. Perhaps something from the first paragraph on details can
be moved to the description
Henrik Bengtsson
# 
A possible solution could be to preserve the original version string
as an attribute, which then as.character() could use, e.g.
List of 1
 $ :Classes 'package_version', 'numeric_version'  hidden list of 1
  ..$ : int [1:3] 1 2 4
 - attr(*, "class")= chr [1:2] "package_version" "numeric_version"
 - attr(*, "source")= chr "1.2-4"

I don't want to divert this "feature request", but a related
discussion is whether we really need to support anything but x.y.z?
Would it make sense to deprecate support for hyphenated version
components? The one argument I recall for supporting hyphens was for
"date" versions, e.g. YYYY-MM-DD, but R has dropped leading zeros for
a while now;
[1] '2026.1.17'

and CRAN doesn't accept them;

* checking CRAN incoming feasibility (4.1s)
   ...
   Version contains leading zeroes (2026-01-17)

/Henrik

PS.I didn't know that
<https://CRAN.R-project.org/package=%s&version=%s> works -it downloads
<package>_<version>.tar.gz. Thanks.
On Sat, Jan 17, 2026 at 4:10?AM Llu?s Revilla <lluis.revilla at gmail.com> wrote:
Josiah Parry
# 
It would be really nice if the full semantic versioning spec was supported
in R. I can understand why only major.minor.patch would be supported for
CRAN. But for an R package in a development stage, it would be wonderful to
support pre-release and build metadata.

Use of `-` in semver to determine pre-release:
https://semver.org/#spec-item-9

The semver package does a pretty decent job of handling the semver spec:

ver_str <- "1.0.0-alpha+commit-hash"
package_version(ver_str)
#> Error: invalid version specification '1.0.0-alpha+commit-hash'

(sv <- semver::parse_version(ver_str))
#> [1] Maj: 1 Min: 0 Pat: 0 Pre: alpha Bld: commit-hash
as.character(sv)
#> [1] "1.0.0-alpha+commit-hash"


On Sat, Jan 17, 2026 at 10:29?AM Henrik Bengtsson <
henrik.bengtsson at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Charles Plessy
    

    

      
      #
    

  


  

      
Le Sat, Jan 17, 2026 at 11:02:12AM -0800, Josiah Parry a ?crit :

            

      
      
      
    

        
In Debian we have long handled this use?case by using a tilde (~) in
version strings?well before semver was formalized. If you like the
semantics of pre?release identifiers but want to avoid changing the
meaning of - in existing R versioning, adopting ~ might be a practical
alternative.

Relevant documentation:
https://www.debian.org/doc/debian-policy/ch-controlfields.html#version

Have a nice week-end,

Charles Plessy

  
  


  


      

    

    

      
      

        


  

        

      Lluís Revilla
    

    

      
      #
    

  


  

      
It is good to see so much interest in version numbers!

R doesn't require Calendar versioning (https://calver.org/), Semantic
Versioning or Debian versions.
The package_version describe the version numbers as:

            

      
      
      
    

        
The R-extension manual describes them in the section "The DESCRIPTION file":

            

      
      
      
    

        
Note that here there are some versions with a leading 0 (which
package_version accepts!).

I think fully supporting any of these versions schemes would involve
changing how R itself resolves dependencies for packages (mainly to
cover the [:alpha:] characters).
This is a much larger patch/project than what I asked for and in any
way R still would need to manipulate (and compre) them easily.

In terms of implementation if the attribute approach is added it could
be used by format.numeric_version.
But I'm not sure if any package relies on this to remove any "-" and
leading 0 present...

            

      
      
      
    

        
function none would be impacted as it is primarily used to enable or
disable features according to the versions of dependencies installed.



Missatge de Charles Plessy <charles-listes+rd at plessy.org> del dia dg.,
18 de gen. 2026 a les 2:40: