stats::wilcox.test vs exactRankTests::wilcox.eact vs coin:: wilcoxsign_test

1. wilcox.test() in the stats package
This function automatically switches to using a Normal
approximation when there are ties in the data:

  wilcox.test(x, y, paired=TRUE)$p.value
#[1] 0.05802402
(You can suppress the warning (due to ties) by specifying
the argument 'exact=FALSE'.)

This function also uses a continuity correction unless
told not to:

  wilcox.test(x, y, paired=TRUE, correct=FALSE)$p.value
#[1] 0.05061243

2. wilcox.exact() in pkg exactRankTests
This function can handle ties (using the "Wilcoxon" method)
with an 'exact' calculation:

  wilcox.exact(x, y, paired=TRUE)$p.value
#[1] 0.0546875

If you want the Normal approximation:

    wilcox.exact(x, y, paired=TRUE, exact=FALSE)$p.value
#[1] 0.05061243  

3. wilcoxsign_test() in pkg coin
This is the most comprehensive of these functions.
It is also the only one that offers the "Pratt" method
of handling ties. It will default to this method and
a Normal approximation:

  pvalue(wilcoxsign_test(x ~ y))
#[1] 0.08143996

  pvalue(wilcoxsign_test(x ~ y, zero.method="Pratt",
         distribution="asympt"))
#[1] 0.08143996

You can get the results from wilcox.exact() with

  pvalue(wilcoxsign_test(x ~ y, zero.method="Wilcoxon",
         distribution="asympt"))
#[1] 0.05061243

and

  pvalue(wilcoxsign_test(x ~ y, zero.method="Wilcoxon",
         dist="exact"))
#[1] 0.0546875


Extract from: https://stat.ethz.ch/pipermail/r-help/2011-April/274931.html