New strategy to retrieve DESCRIPTION if project uses renv #931
Conversation
Previously we'd always restore the packages from the lockfile, which I had expected to be fast since I had expected the packages to be in the cache. However, this doesn't appear to always be the case, so I instead use a simpler approach where I just force the user to resolve the problem. This means that we can always use the `DESCRIPTION` present in the `.libPaths()` making the code much simpler. Fixes #930
|
now it's faster to include the lockfile 😁 > bench::mark(
+ rsconnect::writeManifest(appFiles = c("index.qmd")),
+ rsconnect::writeManifest(appFiles = c("index.qmd", "renv.lock")), iterations = 3
+ )
ℹ Capturing R dependencies with renv
✔ Found 118 dependencies
ℹ Capturing R dependencies from renv.lock
✔ Found 121 dependencies
# A tibble: 2 × 13
expression min median `itr/sec` mem_alloc `gc/sec` n_itr n_gc total_time result memory time gc
<bch:expr> <bch> <bch:> <dbl> <bch:byt> <dbl> <int> <dbl> <bch:tm> <list> <list> <list> <list>
1 "rsconnect::writeManifest(appFiles = c(\"index.qmd\"))" 8.07s 8.49s 0.118 114MB 1.14 3 29 25.4s <NULL> <Rprofmem> <bench_tm> <tibble>
2 "rsconnect::writeManifest(appFiles = c(\"index.qmd\", \"renv.… 6.38s 6.49s 0.155 113MB 0.618 3 12 19.4s <NULL> <Rprofmem> <bench_tm> <tibble>thanks! |
There was a problem hiding this comment.
I will re-test console and push-button deploys from renv projects before merging; the deploy-pane R environment was different than what @hadley was seeing (because I only use the system R library, not a user library).
NEWS.md
Outdated
| @@ -1,5 +1,10 @@ | |||
| # rsconnect (development version) | |||
|
|
|||
| * `deployApp()` and `writeManifest()` now error if your library and lockfile | |||
There was a problem hiding this comment.
renv.lock rather than "lockfile".
"should err" or "should error"? I prefer "err", but others see that as an "error". 😜
There was a problem hiding this comment.
I feel like "err" means to make a mistake; whereas here we are erroring (or more technically correct, throwing an error).
R/bundlePackageRenv.R
Outdated
| lib_dir = c(lib_dir, .libPaths()) | ||
| ) | ||
| if (!snapshot) { | ||
| lib_versions <- unlist(lapply(deps$Package, packageDescription, fields = "Version")) |
There was a problem hiding this comment.
You are about to load all the package descriptions just below; what about validating against deps$description rather than loading the same information twice?
hadley
changed the title
#Conflicts: # NEWS.md # R/bundlePackageRenv.R
|
Confirmed deployment for an renv project from an environment that does not have a user R library. |
Previously we'd always restore the packages from the lockfile, which I had expected to be fast since I had expected the packages to be in the cache. However, this doesn't appear to always be the case, so I instead use a simpler approach where I just force the user to resolve the problem. This means that we can always use the
DESCRIPTIONpresent in the.libPaths()making the code much simpler.Fixes #930
@edavidaja could you please confirm that this fixes your problem?