Update download.R to allow year vectors for baci data

[OlivazShai]

Description of the bug

While downloading BACI data for single years is functional, downloding data from a vector of years raises errors. Say, seting year = 1995:2022 to download the entire series. The result is the following:

Error in utils::download.file(url = path, destfile = temp, method = "curl",  : 
 	 'url' must be a length-one character vector

Causes

The year parameter is used both for data cleaning and for downloading from the sources, when data for each year has a different url. But for Baci data, download is done from a single url, so we don’t need to generate a path for each year.

The download path is first defined as:

path <- param$url

Which would work. But, to account for sources with different urls by year, we have:

if (!is.null(param$year)) {
    path <- path %>%
      stringr::str_replace("\\$year\\$", as.character(param$year))
  }

If the user-provided year is a one-length vector or NULL, nothing changes and the download proceeds. But if we have an n-length vector, path also becomes a vector, with n repetitions of the url.

Then the download operation below raises the error I presented in the beginning:

if (download_method == "curl") {
    utils::download.file(url = path, destfile = temp, method = "curl", quiet = quiet)
  }

Workaround year = NULL fails

The possible workaround of doing year = NULL downloads everything to dat, including files “country_codes_V202401b” and “product_codes_HS92_V202401b”, since the regex only has the term “V202401b” now.

So rbindlist() in baci.R fails, returning:

Error in data.table::rbindlist(.) : 
  Item 29 has 4 columns, inconsistent with item 1 which has 6 columns. To fill missing columns use fill=TRUE.

Solution

I added a new (simple) exception to the definition of the download path in external_download. It ensures 'path' is the single url in 'param$url' when sourcing baci files.

[IgorRigolon]

Very nice, thank you. I'll make some other changes later today as I'm afraid this error might the affecting other functions as well. I added the str_replace bit withiut realising how it interacts with year vectors.

[IgorRigolon]

Pelo que eu testei essa outra forma também resolve. É que alguns links têm tipo www.../...$year$...., e usamos esses placeholders pra substituir o ano automaticamente. Pus um if (str_detect(path, "$year$")) pra só tentar usar o str_replace nesses casos e não gerar esse erro quando year é um vetor.

[OlivazShai]

Muito bom, fica melhor mesmo!