The goal of b64 is to provide a very fast, lightweight, and vectorized base64 encoder and decoder.


You can install the development version of b64 like so:



Encode to base64 using encode().


hello <- encode("Hello, from extendr")
#> [1] "SGVsbG8sIGZyb20gZXh0ZW5kcg=="

Decode using decode()

decoded <- decode(hello)
#> <blob[1]>
#> [1] blob[19 B]

We can convert the decoded base64 to characters and see how it worked.

#> [1] "Hello, from extendr"


Both encode() and decode() are vectorized.

lorem <- unlist(lorem::ipsum(5, 1,  5))
#> [1] "Sit ligula senectus litora viverra consequat."          
#> [2] "Consectetur vulputate vivamus sapien a ridiculus porta."
#> [3] "Ipsum orci cras posuere lacus."                         
#> [4] "Lorem nostra hendrerit nascetur vel duis consequat."    
#> [5] "Adipiscing dui blandit vestibulum bibendum?"

encoded <- encode(lorem)
#> [1] "U2l0IGxpZ3VsYSBzZW5lY3R1cyBsaXRvcmEgdml2ZXJyYSBjb25zZXF1YXQu"                
#> [2] "Q29uc2VjdGV0dXIgdnVscHV0YXRlIHZpdmFtdXMgc2FwaWVuIGEgcmlkaWN1bHVzIHBvcnRhLg=="
#> [3] "SXBzdW0gb3JjaSBjcmFzIHBvc3VlcmUgbGFjdXMu"                                    
#> [4] "TG9yZW0gbm9zdHJhIGhlbmRyZXJpdCBuYXNjZXR1ciB2ZWwgZHVpcyBjb25zZXF1YXQu"        
#> [5] "QWRpcGlzY2luZyBkdWkgYmxhbmRpdCB2ZXN0aWJ1bHVtIGJpYmVuZHVtPw=="

We can decode all of these using decode() as well. This will always return a blob object.

#> <blob[5]>
#> [1] blob[45 B] blob[55 B] blob[30 B] blob[51 B] blob[43 B]

Encoding and decoding files

b64 shines when encoding and decoding files. encode_file() and decode_file() both work by reading a file as a stream making it far faster than the alternative.

tmp <- tempfile() 
fp <- "https://github.com/datablist/sample-csv-files/raw/main/files/leads/leads-100000.csv"

download.file(fp, tmp)

  b64 = encode_file(tmp),
  base64enc = base64enc::base64encode(tmp)
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 b64          39.8ms   41.3ms     24.0       24MB      0  
#> 2 base64enc   112.1ms  115.2ms      8.56    66.5MB     17.1

While the encoding is very impressive, better yet is the decoding performance.

# create a temp file
tmp2 <- tempfile()

# encode it and write to tmep file
encode_file(tmp) |>
  charToRaw() |>

  b64 = decode_file(tmp2),
  base64enc = base64enc::base64decode(file(tmp2))
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 b64          16.1ms   16.8ms     56.1       18MB     9.34
#> 2 base64enc   209.1ms  210.7ms      4.75      18MB     0

Alternative engines

Out of the box, b64 provides a number of pre-configured engines that can be used. The function engine() allows you to choose one of these different engines For example, engine("url_safe") provides a standard engine that uses a url-safe alphabet with padding.

url_engine <- engine("url_safe")
url_safe_encoded <- encode("\xfa\xec U", url_engine)
#> [1] "-uwgVQ=="

If we try to decode this using the standard engine, we will encounter an error.

#> Error in decode_(what, eng): Invalid byte 45, offset 0.

We can use our new engine to decode it.

decode(url_safe_encoded, url_engine)
#> <blob[1]>
#> [1] blob[4 B]

Custom Engines

We can create custom engines with new_engine(). This allows us to provide our on alphabet and configuration.

We can use one of the many predefined alphabets or create one our selves with new_alphabet(). We can also specify our engine config using new_config() which lets us choose whether or not to pad and how to handle decoding.

my_eng <- new_engine(
  new_config(TRUE, TRUE, "none")

This engine can be used to encode or decode text.

txt <- "lorem ipsum sit dolor amet"

encode(txt, my_eng)
#> [1] "P4xmNKoUOL/nRKoUQqZo64FjP4xm643hNLE="

Compare this to the standard encoder:

#> [1] "bG9yZW0gaXBzdW0gc2l0IGRvbG9yIGFtZXQ="